en

Overview of Japanese LLMs

[ English | Français | 日本語 ]

📖 Please visit the more readable Web version

The content of this README is available in a more readable format at llm-jp.github.io/awesome-japanese-llm/en. We recommend viewing the Web version to avoid table display issues and layout problems.

A list of publicly available LLMs trained with a focus on Japanese, along with their evaluation benchmarks, maintained by volunteers from various sources like academic papers and other public resources.

::: warning Caution

1. We can't guarantee the accuracy or completeness of any information here.
2. Some information is based on conjecture and might not reflect your specific use case.
3. While many models are released under permissive licenses like MIT or Apache 2.0, some are subject to more restrictive terms including non-commercial use clauses (e.g CC BY-NC-SA 4.0) or other stipulations. :::

Please point out any errors on the issues page. Feel free to contribute directly with a pull request.

::: details Table of Contents {open} [[toc]] :::

Text Generation Models

For multimodal models, see below.

Models built from scratch

General purpose

	Release Year	Architecture	Max Context Length	Training Data	Developer	License / Terms of Use
Sarashina2-8x70B	2024	MoE (8x70b (465b))	8,192	Sparse Upcycling on Sarashina2 (70B)	SB Intuitions	Sarashina Model NonCommercial License
LLM-jp-3 172B	2024	Llama (172b, 172b-instruct2, 172b-instruct3)	4,096	Pre-training: llm-jp-corpus-v3 (2.1T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, magpie-sft-v1.0, Daring-Anteater, FLAN, ichikara-instruction-format, AutoMultiTurnByCalm3-22B, ramdom-to-fixed-multiturn-Calm3, wizardlm8x22b-logical-math-coding-sft-ja, wizardlm8x22b-logical-math-coding-sft_additional-ja, Synthetic-JP-EN-Coding-Dataset-567k DPO (instruct3 only): aya-ja-evol-inst, ac-self-inst	Research and Development Center for Large Language Models	Pre-trained model: LLM-jp-3 172B Terms of Use Post-trained model: llm-jp-3-172b-instruct3 Terms of Use
LLM-jp-3 172B beta2	2024	Llama (172b-beta2, 172b-beta2-instruct2)	4,096	Pre-training: part of llm-jp-corpus-v3 (1.4T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, magpie-sft-v1.0, Daring-Anteater, FLAN, ichikara-instruction-format, AutoMultiTurnByCalm3-22B, ramdom-to-fixed-multiturn-Calm3, wizardlm8x22b-logical-math-coding-sft-ja, wizardlm8x22b-logical-math-coding-sft_additional-ja, Synthetic-JP-EN-Coding-Dataset-567k	Research and Development Center for Large Language Models	LLM-jp-3 172B beta2 Terms of Use
LLM-jp-3 172B beta1	2024	Llama (172b-beta1, 172b-beta1-instruct)	4,096	Pre-training: part of llm-jp-corpus-v3 (0.7T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, Dolly Dataset, OASST1, OASST2, Aya Dataset, ichikara-instruction-format, Daring-Anteater, FLAN	Research and Development Center for Large Language Models	LLM-jp-3 172B beta1 Terms of Use
LLM-jp-3 172B alpha	2024	Llama (172b-alpha1, 172b-alpha1-instruct, 172b-alpha2, 172b-alpha2-instruct)	4,096	Pre-training: part of llm-jp-corpus-v3 (alpha1: 0.7T tokens, alpha2: 1.4T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, Dolly Dataset, OASST1, OASST2, Aya Dataset, ichikara-instruction-format, Daring-Anteater, FLAN	Research and Development Center for Large Language Models	Apache 2.0
Stockmark-2-100B-Instruct-beta	2025	Llama (100B-Instruct-beta, 100B-Instruct-beta-AWQ)	4,096	Pre-training: 1.5T tokens Instruction Tuning DPO	Stockmark	MIT
Stockmark-100b	2024	Llama (100b, 100b-instruct-v0.1)	4,096	Pre-training: RedPajama, Japanese Wikipedia, Japanese mC4, Japanese CommonCrawl, Japanese Patent, Stockmark Web Corpus (910B tokens) Instruction Tuning (LoRA): ichikara-instruction	Stockmark	MIT
PLaMo-100B-Pretrained	2024	Llama1 (100b)	4,096	Pre-training: Japanese CommonCrawl, RefinedWeb, undisclosed (2.0T tokens)	Preferred Elements (Preferred Networks)	PLaMo Non-Commercial License
LLM-jp-3.1	2025	Llama/MoE (8x13b (73b), 8x13b (73b)-instruct4, 13b, 13b-instruct4, 1.8b, 1.8b-instruct4)	4,096	Pre-training: llm-jp-corpus-v3 (2.5T tokens) Continual pre-training: instruction-response pairs (90B tokens) SFT + DPO	Research and Development Center for Large Language Models	Apache 2.0
LLM-jp-3 MoE	2025	MoE (8x1.8b (9.3b), 8x1.8b (9.3b)-instruct2, 8x1.8b (9.3b)-instruct3, 8x13b (73b), 8x13b (73b)-instruct2, 8x13b (73b)-instruct3)	4,096	Drop-Upcycling on LLM-jp-3 (1.8b, 13b)	Research and Development Center for Large Language Models	Apache 2.0
Sarashina2	2024	Llama (7b, 13b, 70b)	7b, 13b: 4,096 70b: 8,192	Pre-training: Japanese Common Crawl, SlimPajama, StarCoder (2.1T tokens)	SB Intuitions	MIT
Sarashina1	2024	GPT-NeoX (7b, 13b, 65b)	2,048	Pre-training: Japanese Common Crawl (1T tokens)	SB Intuitions	MIT
Tanuki-8×8B	2024	MoE (47b) (v1.0, v1.0-AWQ, v1.0-GPTQ-4bit, v1.0-GPTQ-8bit, v1.0-GGUF)	4,096	Pre-training: various Web & synthetic datasets（1.7T tokens） SFT, DPO: various synthetic datasets 2	Matsuo Lab LLM Development Project	Apache 2.0
LLM-jp-4 32B-A3B	2026	Qwen3 MoE (32b-a3b-base, 32b-a3b-thinking, 32b-a3b-thinking-gguf)	65,536	Pre-training + Mid-training: llm-jp-corpus-v4.1, llm-jp-corpus-midtraining-v2 (11.7T tokens) SFT: llm-jp-4-thinking-sft-data DPO: llm-jp-4-32b-a3b-thinking-dpo-data	Research and Development Center for Large Language Models	Apache 2.0
llm-jp-4-kappa	2026	Qwen3 MoE (32b-a3b-v0.1)	65,536	Post-trained on LLM-jp-4 32B-A3B (thinking) as follows SFT: Nemotron-Cascade-2-SFT-Data, OpenMathReasoning, OpenCodeReasoning, Nemotron-SFT-Competitive-Programming-v2, llm-jp-4-thinking-sft-data, etc. (443,747 samples in total) GRPO (math): 10,999 problems selected from AceReason-Math GRPO (code): approx. 3.1k samples selected from SYNTHETIC-2	Third Intelligence	Apache 2.03
PLaMo 3	2025	Gemma-based architecture (2b-base, 8b-base, 31b-base)	4,096	Pre-training: English, Japanese, Code, Multilingual (2b: 200B tokens, 8b: 1T tokens, 31b: 3T tokens)	Preferred Networks	PLaMo community license
Way-PLaMo-3-8b-chat	2025	PLaMo 3-based (8b-chat)	4,096	Instruction Following SFT: Alpaca (51.7K), Dolly-15k-ja (15K)	Individual (WayBob)	PLaMo community license
CyberAgentLM3 (CALM3)	2024	Llama (22b-chat, 22b-chat-selfimprove-experimental)	16,384	undisclosed (2.0T tokens)	CyberAgent	Apache 2.0
LLM-jp-3 13B instruct3	2025	Llama (150m, 150m-instruct2, 150m-instruct3, 440m, 440m-instruct2, 440m-instruct3, 980m, 980m-instruct2, 980m-instruct3, 1.8b-instrcut2, 1.8b-instruct3, 3.7b-instruct2, 3.7b-instruct3, 7.2b-instruct2, 7.2b-instruct3, 13b-instruct2, 13b-instruct3)	4,096	Pre-training: llm-jp-corpus-v3 (2.1T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, magpie-sft-v1.0, Daring-Anteater, FLAN, ichikara-instruction-format, AutoMultiTurnByCalm3-22B, ramdom-to-fixed-multiturn-Calm3, wizardlm8x22b-logical-math-coding-sft-ja, Synthetic-JP-EN-Coding-Dataset-567k DPO (instruct3 only): aya-ja-evol-inst, ac-self-inst	Research and Development Center for Large Language Models	Apache 2.0
LLM-jp-3 13B	2024	Llama (1.8b, 1.8b-instruct, 3.7b, 3.7b-instruct, 7.2b, 7.2b-instruct, 13b, 13b-instruct)	4,096	Pre-training: llm-jp-corpus-v3 (2.1T tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, FLAN, ichikara-instruction-format, AutoMultiTurnByCalm3-22B, ramdom-to-fixed-multiturn-Calm3, wizardlm8x22b-logical-math-coding-sft_additional-ja, Synthetic-JP-EN-Coding-Dataset-567k	Research and Development Center for Large Language Models	Apache 2.0
llm-jp-3-3.7b-instruct-EZO	2024	Llama (3.7b-instruct-EZO-Common, 3.7b-instruct-EZO-Humanities)	4,096	additionally trained on LLM-jp-3 (3.7B)	Axcxept	Apache 2.0
LLM-jp-13B v2.0	2024	Llama (13b-v2.0, 13b-instruct-full-dolly-ichikara_004_001_single-oasst-oasst2-v2.0, 13b-instruct-full-ac_001-dolly-ichikara_004_001_single-oasst-oasst2-v2.0, 13b-instruct-full-ac_001_16x-dolly-ichikara_004_001_single-oasst-oasst2-v2.0)	4,096	Pre-training: llm-jp-corpus-v2 (260B tokens) Instruction Tuning: ichikara-instruction, AnswerCarefully Dataset, Dolly Dataset, OASST1, OASST2	LLM-jp	Apache 2.0
Fugaku-LLM	2024	GPT (13B, 13B-instruct, 13B-instruct-gguf)	2,048	Pre-training: undisclosed dataset Instruction Tuning: OASST1, Dolly Dataset, GSM8K	Titech, Tohoku Univ., Fujitsu, RIKEN, Nagoya Univ., CyberAgent, Kotoba Technologies	Fugaku-LLM Terms of Use
LLM-jp-13B v1.1	2024	GPT (13b-instruct-lora-dolly_en-dolly_ja-ichikara_003_001-oasst_en-oasst_ja-v1.1, 13b-instruct-full-dolly_en-dolly_ja-ichikara_003_001-oasst_en-oasst_ja-v1.1, 13b-dpo-lora-hh_rlhf_ja-v1.1)	2,048	Instruction Tuning (LoRA or Full-parameter FT): Dolly Dataset, OASST1, ichikara-instruction DPO (LoRA): HH RLHF	LLM-jp	Apache 2.0
LLM-jp-13B	2023	GPT (1.3b-v1.0, 13b-v1.0, 13b-instruct-full-jaster-v1.0, 13b-instruct-full-jaster-dolly-oasst-v1.0, 13b-instruct-full-dolly-oasst-v1.0, 13b-instruct-lora-jaster-v1.0, 13b-instruct-lora-jaster-dolly-oasst-v1.0, 13b-instruct-lora-dolly-oasst-v1.0)	2,048	Pre-training: llm-jp-corpus (Wikipedia, Japanese mC4, The Pile, Stack) (300B tokens) Instruction Tuning (Full-parameter FT or LoRA): jaster, Dolly Dataset, OASST1	LLM-jp	Apache 2.0
PLaMo-13B	2023	Llama4 (13b, 13b-instruct, 13b-instruct-nc)	base: 4,096 instruct, instruct-nc: 8,192	Pre-training: C4, Project Gutenberg, RedPajama, Japanese Wikipedia, Japanese mC4 (1.5T tokens) Instruction Tuning: Dolly, HH RLHF, OASST1, wikinews (+Alpaca in NC model)	Preferred Networks	Apache 2.0 (CC BY-NC 4.0 as for NC model)
Stockmark-13b	2023	Llama (13b, 13b-instruct)	2,048	Pre-training: Japanese Wikipedia, Japanese CC-100, Japanese mC4, Japanese CommonCrawl, Japanese Patent, Stockmark Web Corpus (220B tokens) Instruction Tuning (LoRA): ichikara-instruction	Stockmark	base: MIT instruct: CC BY-NC-SA 4.0
Weblab-10B	2023	GPT-NeoX (10b, 10b-instruction-sft)	2,048	Japanese mC4, The Pile (600B tokens) Instruction Tuning: Alpaca, FLAN	University of Tokyo Matsuo-Iwasawa Lab	CC BY‑NC 4.0
LLM-jp-4 8B	2026	Llama (8b-base, 8b-instruct, 8b-thinking, 8b-thinking-gguf)	65,536	Pre-training + Mid-training: llm-jp-corpus-v4.1, llm-jp-corpus-midtraining-v2 (11.7T tokens) SFT: llm-jp-4-thinking-sft-data DPO (thinking only): llm-jp-4-8b-thinking-dpo-data	Research and Development Center for Large Language Models	Apache 2.0
PLaMo 2.1 8B	2025	hybrid architecture like Samba (8b-cpt)	32,768	Training details undisclosed	Preferred Networks	PLaMo community license
PLaMo 2 8B	2025	hybrid architecture like Samba (8b)		mainly Japanese and English data (6T tokens)	Preferred Networks	PLaMo community license
Tanuki-8B	2024	Tanuki (8b) (v1.0, v1.0-AWQ, v1.0-GPTQ-4bit, v1.0-GPTQ-8bit, v1.0-GGUF)	4,096	Pre-training: various Web & synthetic datasets（1.3T tokens） SFT, DPO: various synthetic datasets 2	Matsuo Lab LLM Development Project	Apache 2.0
Japanese StableLM Alpha	2023	GPT-NeoX (base-alpha-7b, instruct-alpha-7b, instruct-alpha-7b-v2)	2,048	Wikipedia, Japanese CC‑100, Japanese mC4, Japanese OSCAR, RedPajama, private datasets5 (750B tokens) Instruction Tuning: Dolly, HH‑RLHF, wikinews, Alpaca (discarded in v2)	Stability AI	base: Apache 2.0 instruct (v1): Research license instruct (v2): Apache 2.0
CyberAgentLM2 (CALM2)	2023	Llama (7b, 7b-chat, 7b-chat-dpo-experimental)	base: 4,096 chat: 32,768	publicly available Japanese and English datasets (details unknown) (1.3T tokens) DPO: Chatbot Arena Conversations JA (calm2) Dataset	CyberAgent	Apache 2.0 (CC BY 4.0 as for DPO model)
OpenCALM	2023	GPT-NeoX (small, medium, large, 1b(1.4b), 3b(2.7b), 7b(6.8b))	2,048	Japanese Wikipedia, Japanese mC4, Japanese CC‑100	CyberAgent	CC BY‑SA 4.0
Stormy	2023	GPT-NeoX (7b(6.8b))	2,048	OpenCALM fine-tuned on llm-japanese-dataset v0 non-translation tasks	University of Tokyo Izumi Lab	CC BY‑SA 4.0
ByGPT-JP	2025	Llama-based (multi-lm-head-6.5b-alpha)	5,760	Subset of llm-jp-corpus-v3 (ja_cc, ja_warp_html, ja_warp_pdf, ja_wiki, kaken)	Tohoku University NLP Group	Apache 2.0
rinna GPT (En-Ja Bilingual)	2023	GPT-NeoX (4b(3.8b), 4b(3.8b)-8k, 4b(3.8b)-instruction-sft, 4b(3.8b)-instruction-ppo)	8k model: 8,192 others: 2,048	Wikipedia, Japanese CC‑100, Japanese C4, RedPajama, The Pile (524B tokens) Instruction Tuning: HH‑RLHF, FLAN PPO: HH‑RLHF for reinforcement learning 8k: trained with long context	rinna	MIT
japanese-large-lm	2023	GPT-NeoX (1.7b, 3.6b, 1.7b-instruction-sft, 3.6b-instruction-sft)	2,048	Japanese Wikipedia, Japanese CC‑100, Japanese C4, Japanese OSCAR and private datasets (650GB) Instruction Tuning: OASST1	LINE	Apache 2.0
rinna GPT (Japanese only)	2023	GPT / GPT-NeoX (xsmall, small, medium, 1b, neox-small, neox-3.6b-instruction-sft-v2, neox-3.6b-instruction-ppo)	≤ 2,048	Japanese Wikipedia, Japanese CC‑100 (1b and later models add Japanese mC4) Instruction Tuning: HH‑RLHF, FLAN, SHP PPO: HH‑RLHF for reinforcement learning	rinna	MIT
Sarashina2.2	2025	Llama (0.5b, 0.5b-instruct-v0.1, 1b, 1b-instruct-v0.1, 3b, 3b-instruct-v0.1)	8,192		SB Intuitions	MIT
RetrievaT5	2023	T5 (small (short), small (medium), small (long), base (short), base (medium), base (long), large (short), large (medium), large (long), xl(3b))		Japanese Wikipedia, Japanese mC4	Retrieva	CC BY‑SA 4.0
Spiral-RetNet-3b-base	2024	RetNet (3b)	2,048	Wikipedia, Japanese CC-100, CulturaX	Spiral.AI	MIT
kotomamba-2.8B	2024	Mamba (2.8B-v1.0)	2,048	Japanese Wikipedia, Swallow Corpus, SlimPajama	Kotoba Technologies	Apache 2.0
ABEJA GPT	2022	GPT / GPT-NeoX (large, neox-2.7b)		Japanese Wikipedia, Japanese CC‑100, Japanese OSCAR	ABEJA	MIT
PLaMo 2.1 2B	2025	Causal decoder-only transformer (2b-cpt)	32,768	Training details undisclosed	Preferred Networks	PLaMo community license
Rakuten AI 2.0 mini	2025	Mistral (mini(1.5b), mini(1.5b)-instruct)	131,072		Rakuten	Apache 2.0
WasedaGPT	2022	GPT (small, xl(1.5b))		Japanese Wikipedia, Japanese CC‑100	Waseda Kawahara Lab	CC BY‑SA 4.0
StockmarkGPT	2023	GPT-NeoX (1.4b)		Japanese Wikipedia (0.88B tokens), Japanese CC‑100 (10.5B tokens), private data (8.6B tokens)	Stockmark	MIT
YellowbackGPT	2021	GPT-NeoX (1.3b)		Japanese Wikipedia, Japanese CC‑100, Japanese OSCAR	Yellowback	Apache 2.0
PLaMo 2 1B	2025	hybrid architecture like Samba (1b)		mainly Japanese and English data (4T tokens)	Preferred Elements (Preferred Networks)	Apache 2.0
Sarashina2.1-1B	2024	Llama (1b)	8,192	Japanese and English data on the web (10T tokens)	SB Intuitions	Sarashina Model NonCommercial License
colorfulscoop GPT	2021	GPT (small)		Japanese Wikipedia	Colorful Scoop	CC BY‑SA 3.0
TitechGPT	2023	GPT (medium, medium-reversed) 6		Japanese Wikipedia, Japanese CC‑100	Titech Okazaki Lab	CC BY‑SA 4.0
KyotoUniversityGPT	2022	GPT (small, medium, large)		Japanese Wikipedia (3.2GB), Japanese CC‑100 (85GB), Japanese OSCAR (54GB)	Kyoto University Language Media Processing Lab	CC BY‑SA 4.0
JapaneseBART	2023	BART (base, large)		Japanese Wikipedia (18M sentences)	Kyoto University Language Media Processing Lab	CC BY‑SA 4.0
Megagon Labs T5	2021	T5 (base)		Japanese mC4 (782 GB), Japanese wiki40b (2 GB)	Megagon Labs (Recruit Co.,Ltd.)	Apache 2.0

Domain Specific

	Release Year	Domain	Architecture	Training Data	Developer	License
SIP-med-LLM/SIP-jmed-llm-3-8x13b-OP-32k-R0.1	2026	Medical	MoE	Continual pre-training on a medical corpus (78.3B tokens) added to LLM-jp-3 MoE (8x13b), context length extended to 32k, followed by Instruction Tuning	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	Apache 2.07
SIP-med-LLM/SIP-jmed-llm-3-8x13b-AC-32k-instruct	2025	Medical	MoE	Continual pre-training on a medical corpus (78.3B tokens) added to LLM-jp-3 MoE (8x13b), context length extended to 32k, followed by Instruction Tuning using PMC-Patients and others	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	SIP-jmed-llm-3-8x13b-AC-32k-instruct Terms of Use7
SIP-med-LLM/SIP-jmed-llm-3-8x13b-OP-4k-base	2025	Medical	MoE	Continual pre-training on a medical corpus (78.3B tokens) added to LLM-jp-3 MoE (8x13b)	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	Apache 2.07
SIP-med-LLM/SIP-jmed-llm-2-8x13b-OP-instruct	2025	Medical	MoE	Pre-trained on a medical corpus (44.2B tokens) added to LLM-jp-3 MoE (8x13b), followed by Instruction Tuning	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	Apache 2.07
SIP-med-LLM/SIP-jmed-llm-3-13b-OP-32k-R0.1	2026	Medical	Llama	Continual pre-training on a medical corpus (78.3B tokens) added to LLM-jp-3.1 (13b), context length extended to 32k, followed by Instruction Tuning	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	Apache 2.07
SIP-med-LLM/SIP-jmed-llm-3-13b-OP-4k-base	2025	Medical	Llama	Continual pre-training on a medical corpus (78.3B tokens) added to LLM-jp-3.1 (13b)	Strategic Innovation Promotion Program (SIP) Phase 3 Project "Generative AI Utilization in the Construction of Integrated Healthcare Systems" Theme 1 "Development and Social Implementation of Open Medical LLM with Safety and Reliability" Research Group	Apache 2.07
AscleLM-1-10B	2026	Medical	Mamba-2 Hybrid (based on Nemotron-H)	Pre-training: Nemotron-CC v1 (approx. 6.3T tokens), open-source Japanese/English/Chinese corpora, code, math, and reasoning data (approx. 8.5T tokens in total), and synthetic data derived from medical papers, medical textbooks, clinical guidelines, and exam questions	University of Tokyo Matsuo-Iwasawa Lab	AscleLM-1-10B Terms of Use8
Japanese Dialog Transformer	2021	Dialog	Transformer	Twitter Japanese reply pairs	NTT	Evaluation Licence
Japanese News BART	2023	Business	BART (base)	Japanese business news articles (21M articles)	Stockmark	MIT
AcademicBART	2023	Science	BART (base)	CiNii Japanese Papers	Ehime University AI Lab	Apache 2.0

Models built off non-Japanese LLMs (w/ continual pre-training on Japanese)

*Includes models that underwent post-training (SFT, DPO, RL, etc.) after continual pre-training.

General purpose

	Release Year	Base Model	Training Data	Developer	License / Terms of Use
GPT-OSS Swallow 120B (120B-SFT-v0.1, 120B-RL-v0.1, 120B-RL-v0.1-MXFP4)	2026	GPT-OSS (120b)	Pre-training: Wikipedia, Swallow Corpus v3.2, Nemotron-CC, Cosmopedia, Laboro ParaCorpus, Swallow Math v2, Swallow Code v2 (419.4B tokens) SFT: GPT-OSS-LMSYS-Chat-1M-Synth, Swallow-Nemotron-Post-Training-Dataset-v1 RL: allenai/Dolci-Think-RL-7B (Math subset)	Swallow Project	Apache 2.0
Llama 3.3 Swallow 70B (70B-v0.4, 70B-Instruct-v0.4)	2025	Llama 3.3 (70b)	Pre-training: Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus, FineMath-4+, Swallow Code Version 0.3 Instruction Tuning: Gemma-2-LMSYS-Chat-1M-Synth, Swallow-Magpie-Ultra-v0.1, Swallow-Gemma-Magpie-v0.1, Swallow-Code-v0.3-Instruct-style	Swallow Project	Llama 3.3 Community License & Gemma Terms of Use
Llama 3.1 Swallow 70B (70B-v0.1, 70B-Instruct-v0.1, 70B-Instruct-v0.3)	2024	Llama 3.1 (70b)	Pre-training: The Stack v2, Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus Instruction Tuning: lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions, lmsys-chat-1m-synth-en-wo-pii-and-template-instructions, filtered-magpie-ultra-ja, filtered-magpie-ultra-en, gemma-magpie	Swallow Project	Llama 3.1 Community License (Gemma Terms of Use is also applied to the Instruct model)
cyberagent/Llama-3.1-70B-Japanese-Instruct-2407	2024	Llama 3.1 (70b)	undisclosed	CyberAgent	Llama 3.1 Community License
Llama 3 Swallow 70B (70B-v0.1, 70B-Instruct-v0.1)	2024	Llama 3 (70b)	Pre-training: Algebraic Stack, Wikipedia, RefinedWeb, Swallow Corpus, Cosmopedia, Laboro ParaCorpus, OpenWebMath Instruction Tuning: OASST1 9	Swallow Project	Llama 3 Community License
turing-motors/Llama-3-heron-brain-70B-v0.3	2024	Llama 3 (70b)	additionally trained on Llama 3 Swallow 70B (details undisclosed)	Turing	Llama 3 Community License
Llama 3 Youko 70B (70b, 70b-instruct, 70b-gptq, 70b-instruct-gptq)	2024	Llama 3 (70b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (5B tokens) Instruction Tuning: undisclosed datasetト10	rinna	Llama 3 Community License
Swallow 70B (70b-hf, 70b-instruct-hf, 70b-instruct-v0.1, 70b-NVE-hf, 70b-NVE-instruct-hf)	2023	Llama 2 (70b)	Pre-training: Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile Instruction Tuning: Dolly Dataset, HH RLHF, OASST1 *v0.1: OASST1, OASST2	Swallow Project	Llama 2 Community License
KARAKURI LM (70b-v0.1, 70b-chat-v0.1)	2024	Llama 2 (70b)	Pre-training: mC4, CC100, OSCAR, RedPajama, undisclosed dataset (16B tokens) SteerLM: OASST2, undisclosed dataset	KARAKURI	Llama 2 Community License11
Japanese Stable LM Beta 70B (base-beta-70b, instruct-beta-70b)	2023	Llama 2 (70b)	Pre-training: Wikipedia, Japanese mC4, Japanese CC-100, Japanese OSCAR, SlimPajama(excluding Books3) (100B tokens) Instruction Tuning: Dolly Dataset, HH RLHF, OASST1	Stability AI	Llama 2 Community License
Fujitsu-LLM-KG (8x7B_cpt, 8x7B_inst-infer_v1, 8x7B_inst-infer_v2, 8x7B_inst-gen_ja, 8x7B_inst-gen_en)	2025	Mixtral-8x7B-Instruct-v0.1 (46.7b)	Pre-training: Knowledge graph parallel corpus (synthesized from Shinra Project, Wikipedia, etc.) 2.1B tokens, total ~300B tokens Instruction Tuning: Knowledge graph reasoning and generation task datasets	Fujitsu	Apache 2.0
Swallow-MX 8x7B (8x7b-NVE-v0.1)	2024	Mixtral-8x7B-Instruct-v0.1 (46.7b)	Pre-training: Algebraic Stack, Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile, The Vault	Swallow Project	Apache 2.0
KARAKURI LM 8x7B Instruct v0.1 (8x7b-instruct-v0.1)	2024	Mixtral-8x7B-Instruct-v0.1 (46.7b)	trained Swallow-MX 8x7B on the following datasets: Dolly Dataset, OASST2, HelpSteer, glaive-code-assistant-v3, glaive-function-calling-v2, synthetic_text_to_sql, MetaMathQA, orca-math-word-problems-200k, rag-dataset-12000, rag-hallucination-dataset-1000, undisclosed dataset	KARAKURI	Apache 2.0 (?)12
KARAKURI LM 8x7B Chat v0.1 (8x7b-chat-v0.1)	2024	Mixtral-8x7B-Instruct-v0.1 (46.7b)	trained Swallow-MX 8x7B on OASST2, HelpSteer, and undisclosed datasets using SteerLM	KARAKURI	Apache 2.0
ABEJA-Mixtral-8x7B-japanese (8x7B-v0.1-japanese, 8x7B-Instruct-v0.1-japanese, 8x7B-Instruct-v0.1-japanese-alpha, 8x7B-Instruct-v0.1-japanese-alpha-merged)	2024	Mixtral-8x7B-Instruct-v0.1 (46.7b) *The model without "Instruct" in its name is based on Mixtral-8x7B-v0.1	Pre-training: Japanese CC, Redpajama, undisclosed dataset （450B tokens）	ABEJA	Apache 2.0
Qwen3 Swallow 32B (32B-CPT-v0.2, 32B-SFT-v0.2, 32B-RL-v0.2, 32B-RL-v0.2-AWQ-INT4)	2026	Qwen3 (32b)	Pre-training: Wikipedia, Swallow Corpus v3.2, Nemotron-CC, Cosmopedia, Laboro ParaCorpus, Swallow Math v2, Swallow Code v2 (209.7B tokens) SFT: GPT-OSS-LMSYS-Chat-1M-Synth, Swallow-Nemotron-Post-Training-Dataset-v1 RL: allenai/Dolci-Think-RL-7B (Math subset)	Swallow Project	Apache 2.0
ELYZA-Thinking-1.0-Qwen-32B (32B)	2025	Qwen 2.5 (32b)	Pre-training + SFT (Reasoning)	ELYZA	Apache 2.0
ELYZA-Shortcut-1.0-Qwen-32B (32B)	2025	Qwen 2.5 (32b)	Pre-training + SFT	ELYZA	Apache 2.0
ABEJA-Qwen2.5-32b-Japanese-v1.0 (v1.0)	2025	Qwen2.5-32B-Instruct (32b)	Continual pre-training + SFT + DPO: ~20,000 synthetic and human-annotated datasets (specialized for extraction and reasoning)	ABEJA	Apache 2.0
Qwen2.5 Bakeneko 32B (qwen2.5-bakeneko-32b, qwen2.5-bakeneko-32b-instruct, deepseek-r1-distill-qwen2.5-bakeneko-32b, qwq-bakeneko-32b)	2025	Qwen 2.5 (32b)		rinna	Apache 2.0
ABEJA-QwQ32b-Reasoning-Japanese-v1.0 (v1.0)	2025	Qwen 2.5 (32b)	ABEJA-Qwen2.5-32b-Japanese-v0.1 + Chat Vector (from QwQ 32b) + continual pre-training	ABEJA	Apache 2.0
ABEJA-Qwen2.5-32b-Japanese-v0.1 (32b-Japanese-v0.1)	2025	Qwen 2.5 (32b)	Pre-training: Common Crawl, Cosmopedia, undisclosed dataset （100B tokens） + Chat Vector	ABEJA	Apache 2.0
neoAI-JP-QwQ-32B (32B)	2025	Qwen 2.5 (32b)	Continual pre-training: ~4B tokens from llm-jp-corpus v3 + Chat Vector (from QwQ-32B)	neoAI	Apache 2.0
neoAI-JP-DeepSeek-Qwen-32B (32B)	2025	Qwen 2.5 (32b)	Continual pre-training: ~4B tokens from llm-jp-corpus v3 + Chat Vector (from DeepSeek-R1-Distill-Qwen-32B)	neoAI	Apache 2.0
Qwen3 Swallow 30B-A3B (30B-A3B-CPT-v0.2, 30B-A3B-SFT-v0.2, 30B-A3B-RL-v0.2, 30B-A3B-RL-v0.2-AWQ-INT4)	2026	Qwen3 (30b-A3B)	Pre-training: Wikipedia, Swallow Corpus v3.2, Nemotron-CC, Cosmopedia, Laboro ParaCorpus, Swallow Math v2, Swallow Code v2 (209.7B tokens) SFT: GPT-OSS-LMSYS-Chat-1M-Synth, Swallow-Nemotron-Post-Training-Dataset-v1 RL: allenai/Dolci-Think-RL-7B (Math subset)	Swallow Project	Apache 2.0
Gemma-2-Llama Swallow 27B (27b-pt-v0.1, 27b-it-v0.1)	2025	Gemma 2 (27b)	Pre-training: Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus, FineMath-4+, Swallow Code Version 0.3 Instruction Tuning: Gemma-2-LMSYS-Chat-1M-Synth, Swallow-Magpie-Ultra-v0.1, Swallow-Gemma-Magpie-v0.1	Swallow Project	Llama 3.3 Community License & Gemma Terms of Use
GPT-OSS Swallow 20B (20B-SFT-v0.1, 20B-RL-v0.1, 20B-RL-v0.1-MXFP4)	2026	GPT-OSS (20b)	Pre-training: Wikipedia, Swallow Corpus v3.2, Nemotron-CC, Cosmopedia, Laboro ParaCorpus, Swallow Math v2, Swallow Code v2 (419.4B tokens) SFT: GPT-OSS-LMSYS-Chat-1M-Synth, Swallow-Nemotron-Post-Training-Dataset-v1 RL: allenai/Dolci-Think-RL-7B (Math subset)	Swallow Project	Apache 2.0
ABEJA-Qwen3-14B-Agentic-256k-v0.1 (v0.1)	2026	Qwen3 (14b)	Continual pre-training: long-context synthetic data (extended context length from 128k to 256k via YaRN) + Reinforcement learning for enhanced agentic capabilities	ABEJA	Apache 2.0
Nekomata 14B (14b, 14b-instruction, 14b-gguf, 14b-instruction-gguf)	2023	Qwen (14b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (66B tokens) Instruction Tuning: Dolly Dataset, FLAN, subsets of llm-japanese-dataset	rinna	Tongyi Qianwen LICENSE
Swallow 13B (13b-hf, 13b-instruct-hf, 13b-instruct-v0.1, 13b-NVE-hf)	2023	Llama 2 (13b)	Pre-training: Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile Instruction Tuning: Dolly Dataset, HH RLHF, OASST1 *v0.1: OASST1, OASST2	Swallow Project	Llama 2 Community License
LEIA-Swallow-13B (13b)	2024	Llama 2 (13b)	additionally trained Swallow 13B using LEIA	Individual (Ikuya Yamada, Ryokan Ri)	Llama 2 Community License
ELYZA-japanese-Llama-2-13b (13b, 13b-instruct, 13b-fast, 13b-fast-instruct)	2023	Llama 2 (13b)	Pre-training: Japanese Wikipedia, Japanese OSCAR, and other crawled data (18B tokens) Instruction Tuning: undisclosed dataset	ELYZA	Llama 2 Community License
cyberagent/Mistral-Nemo-Japanese-Instruct-2408	2024	Mistral NeMo (12b)	undisclosed	CyberAgent	Apache 2.0
NVIDIA-Nemotron-Nano-9B-v2-Japanese (9B)	2026	Nemotron-Nano (9b)	Continual pre-training: Wikipedia, fineweb-2 Japanese, aozorabunko, sip3-ja-general-web-corpus, Nemotron-CC-v2.1, Nemotron-Pretraining-Specialized-v1 SFT: Tool Calling data seeded from Nemotron-Personas-Japan, Nemotron-Post-Training-v3	NVIDIA	NVIDIA Nemotron Open Model License Agreement
Gemma-2-Llama Swallow 9B (9b-pt-v0.1, 9b-it-v0.1)	2025	Gemma 2 (9b)	Pre-training: Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus, FineMath-4+, Swallow Code Version 0.3 Instruction Tuning: Gemma-2-LMSYS-Chat-1M-Synth, Swallow-Magpie-Ultra-v0.1, Swallow-Gemma-Magpie-v0.1	Swallow Project	Llama 3.3 Community License & Gemma Terms of Use
Qwen3 Swallow 8B (8B-CPT-v0.2, 8B-SFT-v0.2, 8B-RL-v0.2, 8B-RL-v0.2-AWQ-INT4)	2026	Qwen3 (8b)	Pre-training: Wikipedia, Swallow Corpus v3.2, Nemotron-CC, Cosmopedia, Laboro ParaCorpus, Swallow Math v2, Swallow Code v2 (209.7B tokens) SFT: GPT-OSS-LMSYS-Chat-1M-Synth, Swallow-Nemotron-Post-Training-Dataset-v1 RL: allenai/Dolci-Think-RL-7B (Math subset)	Swallow Project	Apache 2.0
Llama 3.1 Swallow 8B (8B-v0.1, 8B-Instruct-v0.1, 8B-v0.2, 8B-Instruct-v0.2, 8B-Instruct-v0.3, 8B-Instruct-v0.5)	2025	Llama 3.1 (8b)	Pre-training: The Stack v2, Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus Instruction Tuning: lmsys-chat-1m-synth-ja-wo-pii-and-template-instructions, lmsys-chat-1m-synth-en-wo-pii-and-template-instructions, filtered-magpie-ultra-ja, filtered-magpie-ultra-en, gemma-magpie, Gemma-3-LMSYS-Chat-1M-Synth	Swallow Project	Llama 3.1 Community License (Gemma Terms of Use is also applied to the Instruct model)
Llama 3 Swallow 8B (8B-v0.1, 8B-Instruct-v0.1)	2023	Llama 3 (8b)	Pre-training: Algebraic Stack, Wikipedia, RefinedWeb, Swallow Corpus, Cosmopedia, Laboro ParaCorpus, OpenWebMath Instruction Tuning: OASST1 9	Swallow Project	Llama 3 Community License
turing-motors/Llama-3-heron-brain-8B-v0.3	2024	Llama 3 (8b)	additionally trained on Llama 3 Swallow 8B (details undisclosed)	Turing	Llama 3 Community License
Llama 3 Youko 8B (8b, 8b-instruct, 8b-gptq, 8b-instruct-gptq)	2024	Llama 3 (8b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (22B tokens) Instruction Tuning10: Aya Dataset (Japanese subset), FLAN, Dolly Dataset, HH RLHF, OASST1, OASST2, MetaMathQA, CodeAlpaca Dataset, undisclosed dataset DPO: HelpSteer, HelpSteer2, undisclosed dataset	rinna	Llama 3 Community License
Llama 3 ELYZA JP 8B (8B, 8B-GGUF, 8B-AWQ)	2024	Llama 3 (8b)	undisclosed	ELYZA	Llama 3 Community License
Llama 3 neoAI 8B Chat v0.1 (8B-Chat-v0.1)	2024	Llama 3 (8b)	undisclosed	neoAI	Llama 3 Community License
Llama 3 tedllm (v0)	2024	Llama 3 (8b)	Pre-training: Japanese generic corpus	Tokyo Electron Device	Llama 3 Community License
ELYZA-Shortcut-1.0-Qwen-7B (7B)	2025	Qwen 2.5 (7b)	Pre-training + SFT	ELYZA	Apache 2.0
ELYZA-Diffusion-1.0-Dream-7B (Base-7B, Instruct-7B)	2026	Dream (7b)	Pre-training: Japanese text (62B tokens) Instruction Tuning: Japanese instruction data (0.18B tokens)	ELYZA	Apache 2.0
Swallow 7B (7b-hf, 7b-instruct-hf, 7b-instruct-v0.1, 7b-NVE-hf, 7b-NVE-instruct-hf, 7b-plus-hf)	2023	Llama 2 (7b)	Pre-training: Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile Instruction Tuning: Dolly Dataset, HH RLHF, OASST1 *v0.1: OASST1, OASST2	Swallow Project	Llama 2 Community License
LEIA-Swallow-7B (7b)	2024	Llama 2 (7b)	additionally trained Swallow 7B using LEIA	Individual (Ikuya Yamada, Ryokan Ri)	Llama 2 Community License
ELYZA-japanese-Llama-2-7b (7b, 7b-instruct, 7b-fast, 7b-fast-instruct)	2023	Llama 2 (7b)	Pre-training: Japanese Wikipedia, Japanese OSCAR, and other crawled data (18B tokens) Instruction Tuning: undisclosed dataset	ELYZA	Llama 2 Community License
Youri 7B (7b, 7b-instruction, 7b-chat, 7b-gptq, 7b-instruction-gptq, 7b-chat-gptq)	2023	Llama 2 (7b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (40B tokens) Instruction Tuning: Dolly Dataset, FLAN, subsets of llm-japanese-dataset	rinna	Llama 2 Community License
houou-7b (instruction-7b-v1, instruction-7b-v2, instruction-7b-v3)	2023	Llama 2 (7b)	Instruction-tuned Youri 7B (base) on ichikara-instruction	MoneyForward	Llama 2 Community License
Japanese Stable LM Beta 7B (base-beta-7b, base-ja_vocab-beta-7b, instruct-beta-7b, instruct-ja_vocab-beta-7b)	2023	Llama 2 (7b)	Pre-training: Wikipedia, Japanese mC4, Japanese CC-100, Japanese OSCAR, SlimPajama(excluding Books3) (100B tokens) Instruction Tuning: Dolly Dataset, HH RLHF, OASST1	Stability AI	Llama 2 Community License
SambaLingo-Japanese (Base, Chat)	2024	Llama 2 (7b)	Pre-training: CulturaX Instruction Tuning: ultrachat_200k DPO: ultrafeedback, cai-conversation-harmless	SambaNova Systems	Llama 2 Community License (?)12
blue-lizard (blue-lizard)	2024	Llama 2 (7b)	undisclosed	Deepreneur	Llama 2 Community License
Swallow-MS 7B (7b-v0.1, 7b-instruct-v0.1)	2024	Mistral-7B-v0.1 (7b)	Pre-training: Algebraic Stack, Japanese Wikipedia, RefinedWeb, Swallow Corpus, The Pile Instruction Tuning: Dolly Dataset, OASST1	Swallow Project	Apache 2.0
Rakuten AI 2.0 (8x7B, 8x7B-instruct)	2025	Mistral-7B-v0.1 (7b)		Rakuten	Apache 2.0
RakutenAI-7B (7B, 7B-instruct, 7B-chat)	2024	Mistral-7B-v0.1 (7b)	Pre-training: undisclosed Instruction Tuning: Dolly Dataset, OASST1, datasets converted from the train split of NLU datasets (like jaster), undisclosed dataset	Rakuten	Apache 2.0
Japanese Stable LM Gamma 7B (base-gamma-7b, instruct-gamma-7b)	2023	Mistral-7B-v0.1 (7b)	Pre-training: Wikipedia, Japanese mC4, Japanese CC-100, Japanese OSCAR, SlimPajama(excluding Books3) (100B tokens) Instruction Tuning: Dolly Dataset, HH RLHF, wikinews subset of llm-japanese-dataset	Stability AI	Apache 2.0
ChatNTQ JA 7B (7b-v1.0)	2024	Mistral-7B-v0.1 (7b)	Instruction-tuned Japanese Stable LM Gamma 7B (base) on their own datasets	NTQ Solution	Apache 2.0
Shisa Gamma 7B (7b-v1)	2023	Mistral-7B-v0.1 (7b)	Instruction-tuned Japanese Stable LM Gamma 7B (base) on ultra-orca-boros-en-ja	AUGMXNT	Apache 2.0 (?)12
Shisa 7B (base-7b-v1, 7b-v1)	2023	Mistral-7B-v0.1 (7b)	Pre-training: shisa-pretrain-en-ja-v1 (8B tokens) Instruction Tuning & DPO: ultra-orca-boros-en-ja, shisa-en-ja-dpo-v1	AUGMXNT	Apache 2.0 (?)12
Karasu (7B, 7B-chat, 7B-chat-plus, 7B-chat-plus-unleashed)	2024	Mistral-7B-v0.1 (7b)	Additionally trained Shisa 7B (base) on Aozora Bunko, Japanese Law Precedent Dataset, Japanese Wikipedia, Japanese domain webscrapes from the Japanese subset of CulturaX, UltraChat 200k (7B tokens) Instruction Tuning: ultra-orca-boros-en-ja-v1, OASST1, ShareGPT, undisclosed dataset	Lightblue	Apache 2.0 (?)12
Nekomata 7B (7b, 7b-instruction, 7b-gguf, 7b-instruction-gguf)	2023	Qwen (7b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (66B tokens) Instruction Tuning: Dolly Dataset, FLAN, subsets of llm-japanese-dataset	rinna	Tongyi Qianwen LICENSE
lightblue/japanese-mpt-7b	2023	MPT (7b)	Japanese mC4	Lightblue	Apache 2.0
Japanese Stable LM 3B-4E1T (3b-4e1t-base, 3b-4e1t-instruct)	2024	StableLM-3B-4E1T (3b)	Pre-training: Wikipedia, Japanese mC4, Japanese CC-100, Japanese OSCAR, SlimPajama(excluding Books3) (100B tokens) Instruction Tuning: Dolly Dataset, HH RLHF, wikinews subset of llm-japanese-dataset	Stability AI	Apache 2.0
kotomamba-2.8B-CL	2024	mamba-2.8b-slimpj (2.8b)	Japanese Wikipedia, Swallow Corpus, SlimPajama	Kotoba Technologies	Apache 2.0
Gemma-2-Llama Swallow 2B (2b-pt-v0.1, 2b-it-v0.1)	2025	Gemma 2 (2b)	Pre-training: Wikipedia, DCLM-baseline-1.0, Swallow Corpus Version 2, Cosmopedia, Laboro ParaCorpus, FineMath-4+, Swallow Code Version 0.3 Instruction Tuning: Gemma-2-LMSYS-Chat-1M-Synth, Swallow-Magpie-Ultra-v0.1, Swallow-Gemma-Magpie-v0.1	Swallow Project	Llama 3.3 Community License & Gemma Terms of Use
Gemma 2 Baku 2B (2b, 2b-it)	2024	Gemma 2 (2b)	Pre-training: Wikipedia, Japanese C4, Japanese CC-100, Japanese OSCAR, The Pile, undisclosed dataset (80B tokens) OPRO: undisclosed dataset 13	rinna	Gemma Terms of Use
Japanese Stable LM 2 1.6B (base, instruct)	2024	Stable LM 2 1.6B (1.6b)	Pre-training: Wikipedia, CulturaX Instruction Tuning: jaster, ichikara-instruction, alpaca-gpt4-japanese, ultra-orca-boros-en-ja-v1	Stability AI	STABILITY AI NON-COMMERCIAL RESEARCH COMMUNITY LICENSE
TinySwallow-1.5B (1.5B, 1.5B-Instruct, 1.5B-Instruct-q4f32_1-MLC, 1.5B-Insturct-GGUF)	2025	Qwen2.5 (1.5b)	Pre-training: trained using the TAID method (with Qwen2.5 (32b) as the teacher model) Instruction Tuning: Gemma-2-LMSYS-Chat-1M-Synth, swallow-magpie-ultra-v0.1, swallow-gemma-magpie-v0.1	Sakana AI, Swallow Project	Apache 2.0
EQUES/OpenRS3-GRPO-ja	2025	Qwen2.5 (1.5b)	GRPO training on TinySwallow-1.5B-Instruct with kunishou/OpenMathInstruct-1-1.8m-ja	EQUES Inc.	?
EQUES/TinyDeepSeek-JP-1.5B	2025	Qwen2.5 (1.5b)	TAID distillation on TinySwallow-1.5B-Instruct with EQUES/japanese_ultrachat_6.6k	EQUES Inc.	Apache 2.0
EQUES/TinySwallow-Stratos-1.5B	2025	Qwen2.5 (1.5b)	Reasoning enhancement on TinySwallow-1.5B-Instruct with Bespoke-Stratos-35k	EQUES Inc.	Apache 2.0
karasu-1.1B	2023	TinyLlama (1.1b)	Pre-training: Japanese OSCAR, Japanese mC4 (3B tokens)	Lightblue	Apache 2.0

Domain specific

	Release Year	Domain	Base Model	Training Data	Developer	License
Weblab-MedLLM-GLM-4.7	2026	Medicine	GLM-4.7 (355b-A32B)	Continual pre-training on proprietary synthetic data derived from medical papers, medical textbooks, clinical guidelines, and exam questions, followed by post-training	University of Tokyo Matsuo-Iwasawa Lab	MIT14
Weblab-MedLLM-Qwen3-235B (Instruct, Thinking)	2026	Medicine	Qwen3 (235b-A22B)	Continual pre-training on proprietary synthetic data derived from medical papers, medical textbooks, clinical guidelines, and exam questions, followed by post-training	University of Tokyo Matsuo-Iwasawa Lab	Apache 2.014
Weblab-MedLLM-gpt-oss-120b	2026	Medicine	GPT-OSS (120b)	Continual pre-training on proprietary synthetic data derived from medical papers, medical textbooks, clinical guidelines, and exam questions, followed by post-training	University of Tokyo Matsuo-Iwasawa Lab	Apache 2.014
Medical-GPT-OSS-Swallow-120B	2026	Medicine	GPT-OSS (120b)	Continual pre-training of GPT-OSS Swallow 120B (RL) on a mixture emphasizing medical-domain text (biomedical literature, medical synthetic data, medical QA, clinical guideline-style text)	Swallow Project	Apache 2.07
pfnet/Preferred-MedLLM-Qwen-72B	2025	Medicine	Qwen2.5 (72b)	Continual pre-training on a proprietary corpus of medical-related text	Preferred Networks	Qwen LICENSE
Llama3-Preferred-MedSwallow-70B (70B)	2024	Medicine	Llama 3 (70b)	Trained on a proprietary PFN medical dataset (e.g., explanations from pre-2017 Japanese national medical licensing exams)	Preferred Networks	Llama 3 Community License
AIgroup-CVM-utokyohospital/MedSwallow-70b	2024	Medicine	Llama 2 (70b)	Instruction tuning on a Japanese-translated USMLE dataset	University of Tokyo Hospital Department of Cardiovascular Medicine AI Group	CC BY-NC-SA 4.0
Medical-Qwen3-Swallow-32B	2026	Medicine	Qwen3 (32b)	Continual pre-training of Qwen3 Swallow 32B (RL) on a mixture emphasizing medical-domain text (biomedical literature, medical synthetic data, medical QA, clinical guideline-style text)	Swallow Project	Apache 2.07
Medical-Qwen3-Swallow-30B-A3B	2026	Medicine	Qwen3 (30b-A3B)	Continual pre-training of Qwen3 Swallow 30B-A3B (RL) on a mixture emphasizing medical-domain text (biomedical literature, medical synthetic data, medical QA, clinical guideline-style text)	Swallow Project	Apache 2.07
gpt-oss-20b-Ja-Fin (CPT, Thinking)	2026	Finance	GPT-OSS (20b)	Continual pre-training on a Japanese financial corpus built from Common Crawl and other public sources (domain classification and quality filtering applied)	Nomura Research Institute	Apache 2.0
nekomata-14b-pfn-qfin (qfin, qfin-inst-merge)	2024	Finance	Qwen (14b)	Continual pre-training on a finance-specific corpus (~8.1M documents, ~370M tokens): central bank press conferences and policy-meeting summaries, financial-institution reports, glossaries and corporate information, and Wikipedia finance articles (qfin-inst-merge is merged with the instruct model)	Preferred Networks	Tongyi Qianwen LICENSE
Qwen3-14B-Ja-Fin (CPT, Thinking)	2026	Finance	Qwen3 (14b)	Continual pre-training on a Japanese financial corpus built from Common Crawl and other public sources (domain classification and quality filtering applied)	Nomura Research Institute	Apache 2.0
Watashiha-Llama-2-13B-Ogiri-sft (sft, sft-neuron)	2024	Oogiri	Llama 2 (13b)	Pre-training: Japanese C4, CC-100, OSCAR, Japanese/English Wikipedia, proprietary data (65B tokens) SFT: Ogiri data	Watashiha	Llama 2 Community License
Medical-Qwen3-Swallow-8B	2026	Medicine	Qwen3 (8b)	Continual pre-training of Qwen3 Swallow 8B (RL) on a mixture emphasizing medical-domain text (biomedical literature, medical synthetic data, medical QA, clinical guideline-style text)	Swallow Project	Apache 2.07
MedExamDoc-Llama-3.1-Swallow-8B-Instruct-v0.5	2025	Medicine	Llama 3.1 (8b)	Fine-tuned with QLoRA on JMedBench and KokushiMD-10	Ingenta	Llama 3.1 Community License
Llama 3.1 Future Code Ja 8B	2025	Coding	Llama 3.1 (8b)	Continual pre-training on The Stack v2 (code) and a subset of LLM-jp Corpus v3 (Japanese) — 204.9B code and 85.7B natural-language tokens	Future Corp.	Llama 3.1 Community License
Karamaru (Karamaru-v1)	2025	Edo-period Japanese	Llama 3 (8b)	Continual pre-training on ~25M characters of Edo-period text (~13M human-transcribed: Minna de Honkoku, National Institute of Japanese Literature; ~12M AI-transcribed via the RURI kuzushiji OCR model)	Sakana AI	Llama 3 Community License
JPharmatron (7B-base, 7B)	2025	Pharmaceutical	Qwen2.5 (7b)	Continual pre-training on Japanese pharmaceutical text (2B tokens) and English biomedical text (8B tokens)	EQUES Inc.	CC BY-SA 4.0
ELYZA-japanese-CodeLlama-7b (7b, 7b-instruct)	2023	Coding	Code Llama (7b)	Pre-training: Japanese text from OSCAR, Wikipedia, and proprietary crawl data (18B tokens) instruct: ELYZA's proprietary post-training (undisclosed)	ELYZA	Llama 2 Community License
AIBunCho/japanese-novel-gpt-j-6b	2023	Storytelling	GPT-J (6b)	Additional training on Japanese novel data (the base Japanese GPT-J was pre-trained on CC100 Japanese, Wikipedia, and web data)	Individual (Hiroyuki Osone)	CreativeML OpenRAIL-M License
NovelAI/genji-jp	2022	Storytelling	GPT-J (6b)	Fine-tuned on a proprietary Japanese storytelling dataset	NovelAI	？

Models built off non-Japanese LLMs (post-training only, no continual pre-training or details unknown)

General purpose

	Release Year	Base Model	Training Data	Developer	License / Terms of Use
Rakuten AI 3.0 (RakutenAI-3.0)	2026	DeepSeek-V3 (671b) 15	Undisclosed	Rakuten	Apache 2.0
Llama 3.1 Shisa V2 405B (405b)	2025	Llama 3.1 (405b)	High-quality Japanese datasets with SFT/DPO	Shisa.AI	Llama 3.1 Community License
AXCXEPT/EZO-Qwen2.5-72B-Instruct AXCXEPT/EZO-AutoCoTRAG-Qwen2.5-72B-Instruct_q4	2024	Qwen2.5 (72b)		Axcxept	Qwen License
ao-Karasu (72B)	2024	Qwen1.5 (72b)	ultra-orca-boros-en-ja-v1, OASST1, ShareGPT, Japanese technical blogs, News stories, QA site answers, undisclosed dataset	Lightblue	Tongyi Qianwen LICENSE (?)12
Shisa V2.1 70B (70b)	2025	Llama 3.3 (70b)	Combined SFT/DPO/RL/Model merging	Shisa.AI	Llama 3.3 Community License
shisa-ai/shisa-v2-llama3.3-70b	2025	Llama 3.3 (70b)		Shisa.AI	Llama 3.3 Community License
AXCXEPT/Llama-3.1-70B-EZO-1.1-it	2024	Llama 3.1 (70b)		Axcxept	Llama 3.1 Community License
Llama 3 shisa-v1-llama3-70b (70b)	2024	Llama 3 (70b)	ultra-orca-boros-en-ja-v1	Shisa.AI	Llama 3 Community License (?)12
AIgroup-CVM-utokyohospital/Llama-2-70b-chat-4bit-japanese	2023	Llama 2 (70b)		University of Tokyo Hospital Department of Cardiovascular Medicine AI Group	Llama 2 Community License
doshisha-mil/llama-2-70b-chat-4bit-japanese-v1	2023	Llama 2 (70b)		Doshisha University Media Informatics Lab	？
cyberagent/DeepSeek-R1-Distill-Qwen-32B-Japanese	2025	DeepSeek-R1-Distill-Qwen (32b)		CyberAgent	MIT
Flux-Japanese-Qwen2.5-32B-Instruct-V1.0 (V1.0)	2025	Qwen2.5-32B-Instruct (32b)	Precise-tuning: Pinpointing Japanese knowledge, reasoning, and language circuits to apply adjustments to only 5% of parameters. Created 3 specialized models, then integrated through pinpoint merging	FLUX	Apache 2.0
karakuri-ai/karakuri-lm-32b-thinking-2501-exp	2025	QwQ (32b)		KARAKURI	Apache 2.0
shisa-ai/shisa-v2-qwen2.5-32b	2025	Qwen2.5 (32b)		Shisa.AI	Apache 2.0
AXCXEPT/EZO-Qwen2.5-32B-Instruct AXCXEPT/EZO-AutoCoTRAG-Qwen2.5-32B-Instruct	2024	Qwen2.5 (32b)		Axcxept	Apache 2.0
cyberagent/DeepSeek-R1-Distill-Qwen-14B-Japanese	2025	DeepSeek-R1-Distill-Qwen (14b)		CyberAgent	MIT
Shisa V2.1 14B (14b)	2025	Phi-4 (14b)	Combined SFT/DPO/RL/Model merging	Shisa.AI	MIT
shisa-ai/shisa-v2-unphi4-14b	2025	Phi-4 (14b)		Shisa.AI	MIT
EZO-Phi-4 (phi-4-open-R1-Distill-EZOv1, phi-4-deepseek-R1K-RL-EZO)	2025	Phi-4 (14b)		Axcxept	MIT
Qarasu (14B-chat-plus-unleashed)	2024	Qwen (14b)	ultra-orca-boros-en-ja-v1, OASST1, ShareGPT, undisclosed dataset	Lightblue	Tongyi Qianwen LICENSE (?)12
Sparticle/llama-2-13b-chat-japanese-lora	2023	Llama 2 (13b)		Sparticle	？
izumi-lab/llama-13b-japanese-lora-v0-1ep	2023	Llama (13b)		University of Tokyo Izumi Lab	？
shisa-ai/shisa-v2-mistral-nemo-12b	2025	Mistral NeMo (12b)		Shisa.AI	Apache 2.0
AXCXEPT/EZO-Common-9B-gemma-2-it	2024	Gemma 2 (9b)		Axcxept	Gemma Terms of Use
AXCXEPT/EZO-Humanities-9B-gemma-2-it	2024	Gemma 2 (9b)		Axcxept	Gemma Terms of Use
CAT-Thinking (8B)	2026	Qwen3 (8b)	SFT (math/coding data generated by gpt-oss-120b and translated to Japanese via CAT-Translate-7b) / GRPO	CyberAgent	Apache 2.0
Shisa V2.1 8B (8b)	2025	Qwen3 (8b)	Combined SFT/DPO/RL/Model merging	Shisa.AI	Apache 2.0
AXCXEPT/Qwen3-EZO-8B-beta	2025	Qwen3 (8b)	High-performance reasoning with Deep-Think technique	Axcxept	Apache 2.0
shisa-ai/shisa-v2-llama3.1-8b	2025	Llama 3.1 (8b)		Shisa.AI	Llama 3.1 Community License
AXCXEPT/Llama-3.1-8B-EZO-1.1-it	2024	Llama 3.1 (8b)		Axcxept	Llama 3.1 Community License
Llama 3 Suzume 8B (8B-japanese, 8B-japanese-gguf)	2024	Llama 3 (8b)	megagonlabs/instruction_ja, ShareGPT, undisclosed dataset	Lightblue	Llama 3 Community License (?)12
Llama 3 shisa-v1-llama3-8b (8b)	2024	Llama 3 (8b)	ultra-orca-boros-en-ja-v1	Shisa.AI	Llama 3 Community License (?)12
AXCXEPT/Llama-3-EZO-8b-Common-it	2024	Llama 3 (8b)		Axcxept	Llama 3 Community License
lightblue/DeepSeek-R1-Distill-Qwen-7B-Japanese	2025	DeepSeek-R1-Distill-Qwen (7b)		Lightblue	Apache 2.0
ABEJA-Qwen2.5-7b-Japanese-v0.1 (v0.1)	2025	Qwen 2.5 (7b)		ABEJA	Apache 2.0
shisa-ai/shisa-v2-qwen2.5-7b	2025	Qwen 2.5 (7b)		Shisa.AI	Apache 2.0
Karasu DPO (7B)	2025	Qwen 2.5 (7b)		Lightblue	Apache 2.0
ganchengguang/Yoko-7B-Japanese-v1	2023	Llama 2 (7b)		Yokohama National University Mori Lab	？
Sparticle/llama-2-7b-chat-japanese-lora	2023	Llama 2 (7b)		Sparticle	？
izumi-lab/llama-7b-japanese-lora-v0-5ep	2023	Llama (7b)		University of Tokyo Izumi Lab	？
lightblue/jod	2023	Mistral-7B-SlimOrca (7b)		Lightblue	Apache 2.0
NTQAI/chatntq-7b-jpntuned	2023	RWKV-4 World (7b)		NTQ Solution	？
Qwen3.5-FT-Japanese-CoT-4B	2026	Qwen3.5 (4b)	Undisclosed	Individual (Aname-Tommy)	MIT
Borea (Jp, Common, Coding)	2024	Phi-3.5 (3.8b)		Axcxept	MIT
Shisa V2.1 3B (3b)	2025	Llama 3.2 (3b)	Combined SFT/DPO/RL/Model merging	Shisa.AI	Llama 3.2 Community License
AXCXEPT/EZO-Llama-3.2-3B-Instruct-dpoE	2024	Llama 3.2 (3b)		Axcxept	Llama 3.2 Community License
Gemma-2-JPN (2b-jpn-it)	2024	Gemma 2 (2b)		Google	Gemma Terms of Use
AXCXEPT/EZO-gemma-2-2b-jpn-it	2024	Gemma 2 (2b)		Axcxept	Gemma Terms of Use
AXCXEPT/EZO-Common-T2-2B-gemma-2-it	2024	Gemma 2 (2b)		Axcxept	Gemma Terms of Use
Shisa V2.1 1.2B (1.2b)	2025	LFM2 (1.2b)	Combined SFT/DPO/RL/Model merging	Shisa.AI	LFM Open License v1.0
LFM2.5-1.2B-JP (1.2B-JP, 1.2B-JP-202606)	2026	LFM2.5 (1.2b)	Undisclosed	Liquid AI	LFM Open License v1.0
Qwen3.5-FT-Japanese-CoT-0.8B	2026	Qwen3.5 (0.8b)	Undisclosed	Individual (Aname-Tommy)	MIT

Domain specific

	Release Year	Domain	Base Model	Developer	License
JMedLoRA (llama2-jmedlora-6.89ep)	2023	Medicine	Llama 2 (70b)	University of Tokyo Hospital Department of Cardiovascular Medicine AI Group	CC BY-NC 4.0
pfnet/Qwen3-1.7B-pfn-qfin	2025	Finance	Qwen3 (1.72b)	Preferred Networks	PLaMo Community License
pfnet/Qwen2.5-1.5B-pfn-qfin	2025	Finance	Qwen2.5 (1.54b)	Preferred Networks	PLaMo Community License

Merged models

	Release Year	Original Models (Japanese LLMs in bold)	Developer	License
EQUES/MedLLama3-JP-v2	2024	Llama 3 Swallow 8B (Instruct), OpenBioLLM-8B, MMed-Llama 3 8B, Llama 3 ELYZA JP 8B	EQUES	Llama 3 Community License
EvoLLM-JP-A (v1-7B)	2024	Shisa Gamma 7B (v1), Arithmo2 Mistral 7B, Abel 7B 002	Sakana AI	Apache 2.0
EvoLLM-JP (v1-7B, v1-10B)	2024	Shisa Gamma 7B (v1), WizardMath-7B-V1.1, Abel 7B 002	Sakana AI	MICROSOFT RESEARCH LICENSE
EQUES/TinyQwens-Merge-1.5B	2025	SakanaAI/TinySwallow-1.5B-Instruct, EQUES/TinySwallow-Stratos-1.5B, deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B, Qwen/Qwen2.5-1.5B-Instruct	EQUES Inc.	Apache 2.0

API-based models

	Release Year	Max Context Length	Developer	Platform
PLaMo API	2024	32,768	Preferred Networks	self-owned
AI Novelist	2021	2,400 ~ 8,192	Bit192	self-owned
LHTM-OPT	2024		alt Inc.	AWS Marketplace (SageMaker)
Syn (Syn, Syn Pro)	2025	32,768	KARAKURI, Upstage	AWS Marketplace (SageMaker)
tsuzumi (tsuzumi-7b)	2024		NTT	Microsoft Foundry

Encoder models

General purpose

	Architecture	Max Input Length	Training Data	Developer	License	HuggingFace? 16
ModernBERT-Ja	ModernBERT	8,192	Japanese and English corpora	SB Intuitions	MIT	◯ (30m, 70m, 130m, 310m)
llm-jp-modernbert	ModernBERT	8,192	Japanese subset of llm-jp-corpus-v4 (0.69T tokens)	Research and Development Center for Large Language Models	Apache 2.0	◯
KyotoUniBERT	BERT (base, large)	512	Japanese Wikipedia (18M articles)	Kyoto University Language Media Processing Lab	Apache 2.0	△
TohokuUniversityBERT	BERT (base, large)	512	base (v1): Japanese Wikipedia (17M articles / 2.6GB) base (v2) & large: Japanese Wikipedia 4.0GB base (v3) & large (v2): Japanese Wikipedia (4.9GB), Japanese CC‑100 (74.3GB)	Tohoku University NLP Group	base (v1, v2) & large: CC BY‑SA 3.0 base (v3) & large (v2): Apache 2.0	◯ (base (v1), base (v1, char-level), base (v2), base (v2, char-level), large, large (char-level), base (v3), base (v3, char-level), large (v2), large (v2, char-level))
TohokuNLP BERT-alpha 500M	Llama-based encoder17	4,096 or 8,192	Japanese subset of llm-jp-corpus-v3	Tohoku University NLP Group	Apache 2.0	◯ (sq4096-alpha, sq8192-alpha)
ByBERT-JP	Llama-based encoder17	100m, 200m, 400m: 3,072 v2-100m: 4,096	Japanese subset of llm-jp-corpus-v3 100m: 623B tokens 200m: 637B tokens 400m: 1.23T tokens v2-100m: 2.76T tokens	Tohoku University NLP Group	Apache 2.0	◯ (100m, 200m, 400m, v2-100m)
NICT BERT	BERT (base)	512	Japanese Wikipedia	NICT	CC BY 4.0	△
Laboro BERT	BERT (base, large)	512	Japanese Web Corpus (News and blogs, etc) (12GB)	Laboro.AI	CC BY‑NC 4.0	✕
colorfulscoop BERT	BERT (base)	512	Japanese Wikipedia	Colorful Scoop	CC BY‑SA 3.0	◯
UniversityOfTokyoBERT	BERT (small)	512	Japanese Wikipedia (2.9GB)	University of Tokyo Izumi Lab	CC BY‑SA 4.0	◯
chiTra (Sudachi Transformers)	BERT (base)	512	NINJAL Web Japanese Corpus (148GB)	NINJAL, WAP Tokushima Laboratory of AI and NLP	Apache 2.0	△
ACCMS BERT	BERT (base)	512	Japanese Wikipedia (3.3GB)	Kyoto University ACCMS	CC BY‑SA 4.0	◯
HitachiBERT	BERT (base)	512	Japanese Wikipedia, Japanese CC‑100	Hitachi	CC BY‑NC‑SA 4.0	◯18
RetrievaBERT	BERT 19	2,048	Japanese CommonCrawl, RefinedWeb, Chinese Wikipedia, Korean Wikipedia, The Stack	Retrieva	Apache 2.0	◯
Bandai Namco DistilBERT	DistilBERT	512	(Distillation of TohokuUniversityBERT(base))	Bandai Namco Research	MIT	◯
Laboro DistilBERT	DistilBERT	512	(Distillation of Laboro BERT(base))	Laboro.AI	CC BY‑NC 4.0	◯
LINE DistilBERT	DistilBERT	512	(Distillation of LINE internal BERT model)	LINE	Apache 2.0	◯
rinna RoBERTa	RoBERTa (base)	512	Japanese Wikipedia, Japanese CC‑100	rinna	MIT	◯
WasedaRoBERTa	RoBERTa (base, large)	512	Japanese Wikipedia, Japanese CC‑100	Waseda Kawahara Lab	CC BY‑SA 4.0	◯ (base, large, large (seq512))20
InformatixRoBERTa	RoBERTa (base)	512	Japanese Wikipedia, Web Articles (25GB)	Informatix	Apache 2.0	△
KyotoUniversityRoBERTa	RoBERTa (base, large)	512	Japanese Wikipedia, Japanese CC‑100	Kyoto University Language Media Processing Lab	CC BY‑SA 4.0	◯ (base (char-level), large (char-level))
YokohamaNationalRoBERTa	RoBERTa (base)	512	Japanese Wikipedia (3.45GB)	Yokohama National University Mori Lab	Apache 2.0	◯
Megagon Labs RoBERTa	RoBERTa (base)21	1,282	Japanese mC4 (200M sentences)	Megagon Labs (Recruit Co.,Ltd.)	MIT	◯
ACCMS RoBERTa	RoBERTa (base)	512	Japanese Wikipedia (3.3GB) + Japanese CC‑100 (70GB)	Kyoto University ACCMS	CC BY‑SA 4.0	◯
CinnamonELECTRA	ELECTRA (small)	512	Japanese Wikipedia	Cinnamon	Apache 2.0	◯
Megagon Labs ELECTRA	ELECTRA (base)	512	Japanese mC4 (200M sentences)	Megagon Labs (Recruit Co.,Ltd.)	MIT	◯
UniversityOfTokyoELECTRA	ELECTRA (small, base)	512	Japanese Wikipedia (2.9GB)	University of Tokyo Izumi Lab	CC BY‑SA 4.0	◯ (small, base)
JapaneseRoFormer	RoFormer (base)	512	Japanese Wikipedia (3.45GB)	Yokohama National University Mori Lab	Apache 2.0	◯
JapaneseLUKE	LUKE (base, large)	512	Japanese Wikipedia	Studio Ousia	Apache 2.0	◯ (base, large)
KyotoUniversityDeBERTaV2	DeBERTaV2 (tiny, base, large)	512	Japanese Wikipedia, Japanese CC‑100, Japanese OSCAR (171GB)	Kyoto University Language Media Processing Lab	CC BY‑SA 4.0	◯ (tiny, tiny (char-level), base, large)
KyotoUniversityDeBERTaV3	DeBERTaV3 (base)	512	llm-jp-corpus	Kyoto University Language Media Processing Lab	Apache 2.0	◯
UniversityOfTokyoDeBERTaV2	DeBERTaV2 (small, base)	512	Japanese Wikipedia, Japanese Wikinews, Japanese CC-100, Japanese mC4, Japanese OSCAR	University of Tokyo Izumi Lab	CC BY-SA 4.0	◯ (small, base)
GLOBIS DeBERTaV3	DeBERTaV3 (xsmall, base, large)	512	Wikipedia, WikiBooks, Aozora Bunko, Japanese CC-100, Japanese mC4, Japanese OSCAR	GLOBIS	CC BY-SA 4.0	◯ (xsmall, base, large)
JapaneseBigBird	BigBird (base)	4,096	Japanese Wikipedia, Japanese CC‑100, Japanese OSCAR	Waseda Kawahara Lab	CC BY‑SA 4.0	◯
JapaneseLayoutLM	LayoutLM (base)	512	Pre-trained on Japanese Wikipedia, initialized with TohokuUniversityBERT	The Japan Research Institute, Limited	CC BY-SA 3.0	◯
layoutlmv3-japanese-preview	LayoutLMv3 (base)	512	Initialized with WasedaRoBERTa (base), then pre-trained on ~20M Japanese web pages from NDL WARP and document images (PubLayNet, DocLayNet)	Research and Development Center for Large Language Models	Apache 2.0	◯

Domain Specific

	Domain	Architecture	Training Data	Developer	License	HuggingFace?
JapaneseBlogELECTRA	Colloquial language	ELECTRA (small)	Japanese Blog Corpus (354M sentences)	Kitami Institute of Technology Masui-Ptaszynski Lab	CC BY‑SA 4.0	◯
JapaneseSpokenLanguageBERT	Spoken language	BERT (base)	Additional training for TohokuUniversityBERT using Corpus of Spontaneous Japanese (CSJ) (In the DAPT model, the diet record is also used)	Retrieva	Apache 2.0	◯
AcademicRoBERTa	Science	RoBERTa (base)	CiNii Japanese Papers (6.3M sentences)	Ehime University AI Lab	Apache 2.0	◯
local-politics-BERT	Politics	BERT (base)	Wikipedia, Minutes of the National Diet, Minutes of the Local Assembly	Japanese Local Assembly Minutes Corpus Project	CC BY-SA 4.0	◯ (SC-min, SC-minwiki, SC-2M-wiki, SC-2M-min, SC-2M-minwiki, FP-min, FP-minwiki) 22
UBKE-LUKE	Economics	LUKE (base)	Japanese Wikipedia, Securities Reports, Economic News Articles	Uzabase	CC BY-NC	◯
JapaneseFinancialBERT	Finance	BERT (small, base)23	Japanese Wikipedia, Japanese Financial Corpus (27M sentences/5.2GB)	University of Tokyo Izumi Lab	CC BY‑SA 4.0	◯ (small, base)
JapaneseFinancialELECTRA	Finance	ELECTRA (small)	Japanese Wikipedia (20M sentences/2.9GB), Japanese Financial Corpus (27M sentences/5.2GB)	University of Tokyo Izumi Lab	CC BY‑SA 4.0	◯
JapaneseNewsBERT	Business	BERT (base)	Japanese Business Articles (3M articles)	Stockmark	CC BY 4.0	△
JapaneseNewsXLNet	Business	XLNet (base)	Japanese Business Articles (3M articles)	Stockmark	？	◯ ※ Unofficial release
JapaneseNewsALBERT	Business	ALBERT (base)	Japanese Business Articles (3M articles)	Stockmark	？	△
MinpakuBERT	Cultural Heritage	BERT (base)	Additional training with National Museum of Ethnology's cultural heritage data on top of Tohoku University BERT	University of Hyogo Ohshima Lab	MIT	◯ (minpaku-v1, minpaku-v3, minpaku-v3-no-additional-token)
JPharmaBERT	Pharmacy	BERT (base, large)	Japanese Pharmaceutical Documents (2B tokens) + PubMed English Abstracts (8B tokens) + Multilingual Pharmaceutical Data (1.2B tokens)	EQUES	Unknown	◯ (base, large)
medBERTjp	Medicine	BERT (base)	Japanese Wikipedia, Japanese Medical Corpus ("今日の診療プレミアム/Today's Care Premium" Web Version)	Osaka University Hospital Medical Informatics Lab	CC BY‑NC‑SA 4.0	△
JMedRoBERTa	Medicine	RoBERTa (base)	Japanese Medical Papers (11M sentences/1.8GB)	NII Aizawa Lab	CC BY‑NC‑SA 4.0	◯ (ManbyoWordPiece, SentencePiece)24

Sentence and Document Embeddings ²⁵

Bi-Encoders

Single-representation bi-encoders

	Max Context Length	Developer	License
Ruri-v3 (v3-30m, v3-70m, v3-130m, v3-310m)	8,192	Nagoya University Sasano Group	Apache 2.0
PLaMo-Embedding-1B (1b)	4,096	Preferred Networks	Apache 2.0
Sarashina-Embedding-v2 (v2-1b)	8,192	SB Intuitions	Sarashina Model NonCommercial License
sbintuitions/sarashina-embedding-v1-1b	8,192	SB Intuitions	Sarashina Model NonCommercial License
AMBER (base, large)	512	Retrieva	Apache 2.0
RoSEtta (base-ja)	1,024	PKSHA Technology	Apache 2.0
GLuCoSE v2 (base-ja-v2)	512	PKSHA Technology	Apache 2.0
Ruri (small, base, large, small-v2, base-v2, large-v2)	512	Nagoya University Sasano Group	Apache 2.0
Japanese SimCSE (unsup-simcse-ja-base, unsup-simcse-ja-large, sup-simcse-ja-base, sup-simcse-ja-large)	512	Nagoya University Sasano Group	CC BY-SA 4.0
GLuCoSE (base-ja)	512	PKSHA Technology	Apache 2.0
colorfulscoop/sbert-base-ja		Colorful Scoop	CC BY‑SA 4.0
MU-Kindai/SBERT-JSNLI-base MU-Kindai/SBERT-JSNLI-large		Kindai University	？
MU-Kindai/Japanese-SimCSE-BERT-base-unsup MU-Kindai/Japanese-SimCSE-BERT-large-unsup MU-Kindai/Japanese-SimCSE-RoBERTa-base-unsup MU-Kindai/Japanese-SimCSE-BERT-base-sup MU-Kindai/Japanese-SimCSE-BERT-large-sup		Kindai University	MIT
pkshatech/simcse-ja-bert-base-clcmlp		PKSHA Technology	CC BY‑SA 4.0
MU-Kindai/Japanese-MixCSE-BERT-base MU-Kindai/Japanese-MixCSE-BERT-large		Kindai University	MIT
MU-Kindai/Japanese-DiffCSE-BERT-base		Kindai University	MIT
bclavie/fio-base-japanese-v0.1		Individual (Benjamin Clavié)
cl-nagoya/shioriha-large-pt		Nagoya University Sasano Group

Multi-representation bi-encoders

	Developer	License
JaColBERTv2.5 (JaColBERTv2.4, JaColBERTv2.5)	Answer.AI	MIT
JaColBERTv2 (JaColBERTv2)	Individual (Benjamin Clavié)	MIT
JaColBERT (JaColBERT)	Individual (Benjamin Clavié)	MIT

Cross-Encoders

	Developer	License
Ruri-v3 Reranker (310m)	Nagoya University Sasano Group	Apache 2.0
Ruri-Reranker (stage1-small, stage1-base, stage1-large, small, base, large)	Nagoya University Sasano Group	Apache 2.0
hotchpotch/japanese-reranker-cross-encoder-xsmall-v1 hotchpotch/japanese-reranker-cross-encoder-small-v1 hotchpotch/japanese-reranker-cross-encoder-base-v1 hotchpotch/japanese-reranker-cross-encoder-large-v1 hotchpotch/japanese-bge-reranker-v2-m3-v1	Individual (Yuichi Tateno)	MIT

Vision-Language Models

Text+Image to Text

Models built from scratch

*Includes VLMs newly built by combining an LLM with a vision encoder (the base LLM can be either domestic or non-Japanese).

General purpose

	Release Year	Architecture	Training Data	Developer	License / Terms of Use
Stockmark-2-VL-100B-beta (100B-beta)	2025	LLaVA-OneVision	3-stage training: alignment pre-training, caption expansion, instruction and reasoning fine-tuning Synthetic data: Generated from Qwen2.5-VL-72B	Stockmark	Qwen License
Llama-3.1-70B-Instruct-multimodal-JP-Graph (v0.1)	2025	LLaVA (Llama-3.1-Swallow-70B-Instruct-v0.3 + Qwen2-VL-7B-Instruct)	Over 6 million synthetic visual data specialized for chart and graph understanding (text, pie charts, bar charts, flowcharts, etc.), real data (with FastLabel collaboration)	Ricoh	Llama 3.1 Community License & Gemma Terms of Use & Qwen License & MIT & Apache 2.0
KARAKURI VL (32b-instruct-2507, 32b-thinking-2507-exp)	2025	Vision-Language (based on Qwen2.5-VL-32B)	Custom dataset specialized for Japanese computer use: Japanese computer operation records, Japanese document image QA, visual information interpretation, Japanese OCR, flowchart comprehension 3-stage training: Supervised Fine-Tuning (SFT) + model merging + reinforcement learning *thinking model shows reasoning process explicitly using Chain of Thought (CoT) approach	KARAKURI	Apache 2.0
Heron-NVILA (1B, 2B, 15B, 33B)	2025	NVILA	3-stage training: Alignment (558k Japanese image-text pairs + 595k LLaVA-Pretrain), Pre-training (MOMIJI 13M, Japanese image-text pairs 6M, Japanese interleaved data 2M, coyo-700m 6M, mmc4-core 4M, Wikipedia-ja, LLaVA-Pretrain-JA, STAIR captions), Supervised fine-tuning (LLaVA-instruct-v1.5-en, LLaVA-instruct-ja, Japanese photos conversation, JA-VG-VQA conversation, SynthDog-ja, AI2D, SynthDog-en, Sherlock)	Turing	Apache 2.0 & OpenAI Terms of Use
NABLA-VL (15B)	2025	microsoft/phi-4 + HuggingFaceM4/siglip-so400m-14-980-flash-attn2-navit	Supports single image, multiple images, and video inputs. Training details undisclosed	NABLAS	Apache 2.0
Sarashina2-Vision (8b, 14b)	2025	Sarashina2 + Qwen2-VL + 2-layer MLP	3-stage training: Projector warmup (LLaVA-Pretrain 78M English tokens), Vision encoder pre-training (CC3M, CC12M, llm-jp-japanese-image-text-pairs, internal OCR dataset, internal chart caption synthetic dataset 3.8B Japanese + 7.7B English tokens), Visual instruction tuning (Japanese Visual Genome VQA, OCR-VQA, TextVQA, PlotQA, CLEVR translated, DOCCI translated, internal datasets 2.5B Japanese + 1.0B English tokens)	SB Intuitions	MIT
Asagi (2B, 4B, 8B, 14B)	2025	LLaVA	Newly crawled Japanese website images, existing Japanese datasets, Japanese translations of English datasets ~20M samples (data synthesis using English VLM Phi-3.5-vision-instruct and Japanese LLM CALM3-22B-Chat)	University of Tokyo Machine Intelligence Lab.	Apache 2.0
llava-calm2-siglip (llava-calm2-siglip)	2024	LLaVA	coversational data generated from MS-COCO and VisualGenome	CyberAgent	Apache 2.0
LLM-jp-3 VILA 14B (14b)	2024	LLaVA	Japanese image text pairs, LLaVA-Pretrain, Japanese interleaved data, coyo (subset), mmc4-core (subset), llava-instruct-ja, japanese-photos-conv, ja-vg-vqa, synthdog-ja, LLaVA-1.5 instruction data (subset)	Research and Development Center for Large Language Models	Apache 2.0 & OpenAI Terms of Use
LLM-jp-4-VL 9B beta (9b-beta)	2026	LLaVA (llm-jp-4-8b-instruct + siglip2-so400m-patch16-512 + 2-layer MLP)	Jagle and others, ~33.4M samples / ~180B tokens (Japanese + English)	Research and Development Center for Large Language Models	Apache 2.0
PLaMo 2.1 8B VL (8b-vl)	2026	LLaVA (PLaMo 2.1-8B + siglip2-so400m-patch14-384 + MLP)	2-stage training: Stage 1.0: Image adapter training (web-scale image-caption data with Japanese:English = 75:25), Stage 1.5: Instruction Tuning and Japanese visual adaptation via LoRA (VQA, Visual Grounding, object detection, anomaly detection, factory work understanding)	Preferred Networks	PLaMo community license
Heron (blip-ja-stablelm-base-7b-v0, blip-ja-stablelm-base-7b-v1, blip-ja-stablelm-base-7b-v1-llava-620k, git-ja-stablelm-base-7b-v0, git-ELYZA-fast-7b-v0, git-ja-stablelm-base-7b-v1)	2023	BLIP-2 / GIT	v1: LLaVA-Instruct-150K-JA or LLaVA-Instruct-620K-JA v0: LLaVA-Instruct-150K-JA, Japanese STAIR Captions, Japanese Visual Genome VQA dataset	Turing	CC BY-NC 4.0
Japanese Stable VLM (japanese-stable-vlm)	2023	LLaVA	Japanese CC12M, STAIR Captions, Japanese Visual Genome VQA dataset	Stability AI	STABILITY AI JAPANESE STABLE VLM COMMUNITY LICENSE
Japanese InstructBLIP Alpha (japanese-instructblip-alpha)	2023	InstructBLIP	Japanese CC12M, STAIR Captions, Japanese Visual Genome VQA dataset	Stability AI	JAPANESE STABLELM RESEARCH LICENSE
rinna MiniGPT-4 (bilingual-gpt-neox-4b-minigpt4)	2023	MiniGPT-4	CC12M, COCO 2014, Visual Genome, STAIR Captions, Japanese Visual Genome VQA dataset	rinna	MIT
Sarashina2.2-Vision-3B (3.8b)	2025	Sarashina2.2-3B-Instruct + SigLIP + 2-layer MLP	4-stage training + Post-training: Projector warmup (English image captions), Vision encoder pre-training (Japanese charts, OCR, captions), Full model pre-training (interleaved image-text data), Supervised fine-tuning Post-training: Mixed Preference Optimization (Total: 103B Japanese + 157.1B English tokens)	SB Intuitions	MIT
Jagle-VL 2.2B (2.2b-Jagle, 2.2b-FineVision, 2.2b-Jagle-FineVision)	2026	InternVL3 (Qwen3-1.7B + siglip2-so400m-patch16-512 + 2-layer MLP)	Trained on Jagle (large-scale Japanese multimodal post-training dataset, ~9.2M samples) and FineVision (model name suffixes -Jagle / -FineVision / -Jagle-FineVision indicate which combination is used)	Research and Development Center for Large Language Models	Apache 2.0
PLaMo 2.1 2B VL (2b-vl)	2026	LLaVA (PLaMo 2.1-2B + siglip2-so400m-patch14-384 + MLP)	2-stage training: Stage 1.0: Image adapter training (web-scale image-caption data with Japanese:English = 75:25), Stage 1.5: Instruction Tuning and Japanese visual adaptation via LoRA (VQA, Visual Grounding, object detection, anomaly detection, factory work understanding)	Preferred Networks	PLaMo community license

Domain Specific

	Release Year	Architecture	Domain	Developer	License
Med-Asagi (14b-reasoning_beta)	2026	LLaVA	Medicine	University of Tokyo Machine Intelligence Lab.	CC BY-SA 4.0
watashiha/Watashiha-Llama-2-13B-Ogiri-sft-vlm	2024	LLaVA	Oogiri	Watashiha	Llama 2 Community License

Models built off non-Japanese VLMs

	Release Year	Base Model	Training Data	Developer	License
Stockmark-DocReasoner-Qwen2.5-VL-32B (32B)	2026	Qwen2.5-VL-32B-Instruct	2-stage curriculum learning (SFT): Stage 1: ~1.1M samples (incl. 600K synthetic) for basic document structure understanding, Stage 2: ~1.4M samples (incl. 500K synthetic) for multi-step reasoning and Chain-of-Thought Persona-based synthetic data (leveraging Nemotron-Personas-Japan), rendering with matplotlib/HTML/Plotly/LaTeX/mermaid, quality verification via VLM-as-a-judge	Stockmark	Apache 2.0
AXCXEPT/EZO-InternVL2-26B	2024	InternVL2	-	Axcxept	MIT
KARAKURI VL 2 (8b-thinking-2603)	2026	Qwen3-VL-8B-Thinking	Undisclosed	KARAKURI	Apache 2.0
Qwen-3-VL-Ricoh-8B-20260227 (8B-20260227)	2026	Qwen3-VL-8B-Thinking	Reasoning process via Reinforcement Learning (RL)	Ricoh	Apache 2.0

Merged models

	Release Year	Original Models (Japanese LLMs in bold)	Developer	License
Llama-3-EvoVLM-JP-v2 (v2)	2024	Mantis-8B-SigLIP-Llama-3, Llama-3-ELYZA-JP-8B, Bunny-v1.1-Llama-3-8B-V	Sakana AI	Llama 3 Community License
AXCXEPT/Llama-3-EZO-VLM-1	2024	- (trained from Llama-3-EvoVLM-JP-v2)	Axcxept	Llama 3 Community License
EvoVLM-JP (v1-7B)	2024	Shisa Gamma 7B (v1), LLaVA-1.6-Mistral-7B	Sakana AI	Apache 2.0

Text to Image

General Purpose

	Release Year	Architecture	Training Data	Developer	License
CommonArt β (commonart-beta)	2024	PixArt-Σ	CommonCatalog-cc-by, Megalith-10M, Smithonian Open Access, ArtBench (CC-0 only)	AI Picasso	Apache 2.0
EvoSDXL-JP (v1)	2024	Stable Diffusion	- (merged from several diffusion models, including Japanese Stable Diffusion XL)	Sakana AI	Apache 2.026
Japanese Stable Diffusion XL (japanese-stable-diffusion-xl)	2023	Stable Diffusion	undisclosed	Stability AI	STABILITY AI JAPANESE STABLE DIFFUSION XL COMMUNITY LICENSE
TohokuUniversity Stable Diffusion (base, refiner)	2023	Stable Diffusion	WMT2023 Shared Task English-Japanese parallel corpus, about 13 million captions from laion2B-multi	Tohoku University NLP Group	CreativeML OpenRAIL-M License
rinna Stable Diffusion (japanese-stable-diffusion)	2022	Stable Diffusion	LAION-5B Japanese Subset (100M images)	rinna	CreativeML OpenRAIL-M License

Domain Specific

	Release Year	Architecture	Domain	Developer	License
Evo-Nishikie (v1)	2024	Stable Diffusion (ControlNet)	Ukiyo-e	Sakana AI	Apache 2.026
Evo-Ukiyoe (v1)	2024	Stable Diffusion	Ukiyo-e	Sakana AI	Apache 2.026

Text to Video

	Release Year	Architecture	Training Data	Developer	License
AIdeaLab VideoJP (AIdeaLab-VideoJP)	2025	CogVideoX	Pixabay, FineVideo	AIdeaLab	Apache 2.0

Others

	Release Year	Architecture	Training Data	Developer	License
llm-jp-clip (llm-jp-clip-vit-base-patch16, llm-jp-clip-vit-large-patch14)	2024	CLIP	Translation of about 1.5 billion captions from the English subset of ReLAION-5B	Research and Development Center for Large Language Models	Apache 2.0
LY CLIP (clip-japanese-base, v2)	2025	CLIP	CommonCrawl, CC12M, YFCC100M (v2: ~2B image-text pairs from Common Crawl + knowledge distillation)	LY Corp.	Apache 2.0
Recruit CLIP (japanese-clip-vit-b-32-roberta-base)	2023	CLIP	about 120 million captions from laion2B-multi	Recruit Co.,Ltd.	CC BY-4.0
Japanese Stable CLIP (japanese-stable-clip-vit-l-16)	2023	SigLIP	CC12M translated to Japanese, STAIR Captions	Stability AI	STABILITY AI JAPANESE STABLE CLIP COMMUNITY LICENSE
rinna CLIP (japanese-clip-vit-b-16)	2022	CLIP	CC12M translated to Japanese	rinna	Apache 2.0
rinna CLOOB (japanese-cloob-vit-b-16)	2022	CLOOB	CC12M translated to Japanese	rinna	Apache 2.0
HAKUHODO Technologies CLIP (base, deeper, wider)	2024	CLIP	about 120 million captions from laion2B-multi	HAKUHODO Technologies	CC BY-NC-SA 4.0

Speech-Language Models

Automatic Speech Recognition

	Release Year	Architecture	Training Data	Developer	License
Nue ASR (nue-asr)	2023	Nue ASR (HuBERT + LLM)	ReazonSpeech	rinna	Apache 2.0
Kana-Whisper27 (kana-whisper)	2026	Whisper (large-v3-turbo)	Corpus of Spontaneous Japanese (CSJ)	SB Intuitions	MIT
Kotoba-Whisper (v1.0, v1.0-ggml, v1.0-faster, v1.1, bilingual-v1.0, bilingual-v1.0-ggml, bilingual-v1.0-faster, v2.0, v2.0-ggml, v2.0-faster, v2.1, v2.2)	2024	Distil-Whisper	ReazonSpeech	Kotoba Technologies	Apache 2.0
ReazonSpeech (espnet-v1, espnet-next, espnet-v2, nemo-v2)	2024	ESPnet (Conformer-Transducer) / NeMo (FastConformer-RNNT)	ReazonSpeech	Reazon Holdings	Apache 2.0
Reazon HuBERT ASR (rs35kh, rs35kh-bpe)	2025	HuBERT	ReazonSpeech v2.0	Reazon Holdings	Apache 2.0
Reazon Zipformer ASR (rs35kh, rs35kh-bpe)	2025	Zipformer	ReazonSpeech v2.0	Reazon Holdings	Apache 2.0
Reazon wav2vec 2.0 ASR (base-rs35kh, large-rs35kh)	2024	wav2vec 2.0	ReazonSpeech v2.0	Reazon Holdings	Apache 2.0

Text-to-Speech (TTS)

	Release Year	Architecture	Training Data	Developer	License
Sarashina2.2-TTS (sarashina2.2-tts)	2026	TTS built on Sarashina2.2 (0.5B) (CosyVoice + HiFT-GAN)	Legally obtained audio data (purchased sources, public speech archives, and data collected in compliance with applicable domestic laws)	SB Intuitions	Sarashina Model NonCommercial License
Kotoba-Speech (v0.1)	2024	Transformer	undisclosed	Kotoba Technologies	Apache 2.0

Speech Foundation Models / Spoken Dialogue

	Release Year	Architecture	Training Data	Developer	License
LLM-jp-Moshi-v1 (llm-jp-moshi-v1)	2026	Transformer-based text and speech foundation model (Moshi)	J-CHAT (~69,000 hours), LLM-jp-Zoom1 (~1,000 hours)	大規模言語モデル研究開発センター	Apache 2.0
J-Moshi (j-moshi, j-moshi-ext)	2025	Transformer-based text and speech foundation model (Moshi)	Speech dialogue corpus (J-CHAT, Japanese Callhome, CSJ, travel agency dialogue corpus, proprietary chat dialogue corpus, proprietary consultation dialogue corpus), text dialogue corpus (Japanese PersonaChat, Japanese EmpatheticDialogues, Japanese daily dialogue corpus, RealPersonaChat)	Nagoya University Higashinaka Lab	CC BY-NC 4.0
LFM2.5-Audio-1.5B-JP (1.5B-JP)	2026	LFM2.5-Audio (LFM2 + FastConformer)	Undisclosed	Liquid AI	LFM Open License v1.0

Feature Extraction

	Release Year	Architecture	Training Data	Developer	License
NEST-Ja (0.1b, 0.6b)	2026	NEST (FastConformer)	ReazonSpeech v2.0	SB Intuitions	MIT
Kushinada (base, large)	2025	HuBERT	60k hours of audio extracted from large-scale Japanese TV broadcast audio data	Intelligent Media Processing Research Team, AIST	Apache 2.0
Reazon HuBERT (base-k2)	2025	HuBERT	ReazonSpeech	Reazon Holdings	Apache 2.0
UniversityOfTokyoHuBERT (base-jtube)	2024	HuBERT	JTubeSpeech	University of Tokyo Saruwatari & Takamichi Lab	MIT
rinna HuBERT (base, large)	2023	HuBERT	ReazonSpeech	rinna	Apache 2.0
Izanami (base, large)	2025	wav2vec 2.0	60k hours of audio extracted from large-scale Japanese TV broadcast audio data	Intelligent Media Processing Research Team, AIST	Apache 2.0
Reazon wav2vec 2.0 (base, large)	2024	wav2vec 2.0	ReazonSpeech	Reazon Holdings	Apache 2.0
rinna wav2vec 2.0 (base)	2024	wav2vec 2.0	ReazonSpeech	rinna	Apache 2.0
rinna data2vec Audio (base)	2024	data2vec Audio	ReazonSpeech	rinna	Apache 2.0
Reazon Zipformer (base-k2)	2025	Zipformer	ReazonSpeech	Reazon Holdings	Apache 2.0

Music-Language Models

Music-Text Conversion

	Release Year	Architecture	Training Data	Developer	License
Japanese MULAN (japanese-mulan-base)	2025	MULAN (AST + GLuCoSE)	~20k internal music-text pairs	LY Corporation	Apache 2.0

Evaluation Benchmarks for Japanese LLMs

Hybrid Benchmarks

	Description	Developer
Nejumi LLM Leaderboard4	Comprehensively evaluates the Japanese language capabilities of LLMs across application development (coding, function calling), reasoning abilities (mathematical, logical, abstract reasoning), specialized knowledge, and safety evaluation (instruction following, hallucination suppression). The introduction of high-difficulty benchmarks clarifies performance differences among top-tier models. For more details, see this article.	Weights & Biases
Swallow LLM Leaderboard v2	Conducts a comprehensive evaluation of various LLMs based on three types of tasks: Japanese language understanding and generation tasks, Japanese multi-turn dialogue tasks, and English language understanding and generation tasks. v2 supports reasoning-focused models by adopting zero-shot inference and chain-of-thought prompting, evaluating on more challenging benchmarks (12 total tasks: 6 Japanese, 6 English). Also publishes swallow-evaluation, an evaluation script that integrates and improves existing LLM evaluation tools, plus the newly released swallow-evaluation-instruct for reasoning-type models.	Swallow Project

Traditional Benchmarks based on Natural Language Understanding tasks

	Description	Developer
Open Japanese LLM Leaderboard	Evaluates Japanese language models across 14 categories with 71+ tasks using llm-jp-eval.	LLM-jp, Hugging Face
llm-jp-eval	A tool that evaluates Japanese LLMs automatically across multiple datasets. The complete list of supported datasets can be found here (which also includes tasks such as JNLI and JCommonsenseQA from JGLUE).	LLM-jp
JP Language Model Evaluation Harness	A fork by Stability AI of EleutherAI/lm-evaluation-harness. It is a tool for automatically evaluating Japanese LLMs across multiple datasets. The complete list of supported datasets can be found here (which also includes tasks such as JNLI and JCommonsenseQA from JGLUE).	Stability AI
JGLUE	Japanese version of the GLUE benchmark suite, including the MARC-ja, JCoLA, JSTS, JNLI, JSQuAD, and JCommonsenseQA tasks. JCoLA is by the University of Tokyo's Oseki Lab. See here and here (ja only) for further details about each task.	Waseda University Kawahara Lab and Yahoo
JMMLU	A benchmark constructed as a Japanese version of the MMLU Benchmark, consisting of multiple-choice questions from a wide range of academic fields including natural sciences, humanities, and social sciences. In addition to translating the original MMLU, it features newly added problems based on the unique cultural background of Japan (Japan-specific problems).	Waseda University Kawahara Lab

Benchmarks on open-ended generative tasks

	Description	Developer
llm-jp-judge	An integrated evaluation tool for Japanese LLMs using the LLM-as-a-Judge approach. It evaluates across four categories: Japanese quality (rating accuracy, fluency, detail, and relevance on a 1-5 scale), Japanese safety, MT-Bench (English), and MT-Bench (Japanese). The tool separates generation and evaluation phases and supports multiple inference clients including vLLM, OpenAI API, Azure OpenAI, and AWS Bedrock. Details here.	Research and Development Center for Large Language Models
Japanese MT-bench	The Japanese version of MT-bench asks about multi-turn conversational ability. It includes 80 questions, 10 each, from 8 categories: Writing, Roleplay, Reasoning, Math, Coding, Extraction, STEM, Humanities. Some questions have been modified to fit with Japanese culture during the production of the Japanese version. It also includes a script that performs a 10-level absolute evaluation by GPT-4.	Stability AI
ELYZA-tasks-100	Ranking based on model responses to 100 complex and diverse tasks, including tasks testing summarization, correction, abstraction, induction, and other skills. Uses humans to score the model responses and then ranks models based on their mean scores.	ELYZA
Preferred Generation Benchmark (pfgen-bench)	A benchmark to measure the Japanese language generation ability of LLMs based on 50 common sense questions unique to the Japanese context. It evaluates along three axes: Fluency, Truthfulness, and Helpfulness. The evaluation is conducted without using LLM-as-a-Judge by calculating n-gram or rule-based metrics.	Preferred Elements (Preferred Networks)
Rakuda Benchmark	Ranking based on model answers to 40 open-ended questions on Japanese geography, history, politics, and society. Uses GPT-4 to judge model outputs pairwise, and then ranks models by fitting a Maximum Likelihood Elo/Bradley-Terry model to GPT-4's preferences.	YuzuAI
Japanese Vicuna QA Benchmark	This is the Japanese version of vicuna-blog-eval, which is the predecessor of MT-Bench. It includes 80 questions on general knowledge, role-playing, common sense, Fermi estimation, counterfactual thinking, coding, mathematics, and writing. It also includes a script for automatic evaluation by GPT-4 (win-rate calculation). The leaderboard can be found here.	Kyoto University Language Media Processing Lab
Tengu-Bench	Includes 120 free-form questions from various categories. Categories of questions: table interpretation, logic puzzles, idea generation, function calling, long document summarization (over a thousand tokens), conversation summarization, long document closed QA (over a thousand tokens), honorifics, project creation, math, translation, extraction, ethical control, cost estimation, Japan, chit-chat, puns, formatting, construction, business, legal judgment, politics, hypothetical questions.	Lightblue
Shaberi	A framework that can collectively evaluate the Japanese MT-bench, Rakuda Benchmark, ELYZA-tasks-100, and Tengu-Bench. There is also a fork by Shisa.AI.	Lightblue

Benchmarks for measuring performance in specific domains

	Description	Developer
Japanese Language Model Financial Evaluation Harness	A benchmark for Japanese LLM in the financial sector. It includes tasks such as sentiment analysis in finance (chabsa), basic knowledge tasks in securities analysis (cma_basics), tasks related to audits in certified public accountant examinations (cpa_audit), multiple choice question tasks in financial planner exams (fp2), and mock exam tasks for securities salespeople exams (security_sales_1). For more details, please see here.	Preferred Networks
pfmt-bench-fin-ja	A benchmark for measuring the generation capabilities of Japanese LLMs in the financial domain.	Preferred Networks
jfinqa	A Japanese financial numerical reasoning QA benchmark. Contains 1,000 numerical reasoning questions extracted from securities reports of 68 companies. Evaluates financial reasoning capabilities including arithmetic operations, ratio calculations, and DuPont analysis. Available on PyPI and HuggingFace.	Individual (ajtgjmdjp)
Stockmark Business Questions	The collection includes 50 questions that probe knowledge on topics such as market trends, current affairs, social issues, and business trends.	Stockmark
JMED-LLM	A dataset for evaluating LLMs in the Japanese medical domain. It compiles previously developed Japanese medical language processing tasks for LLM benchmarking.	NAIST Social Computing Lab.
JMedBench	A benchmark for LLMs in the Japanese medical field. It includes 20 datasets in 5 types of tasks: multi-choice question-answering, machine translation, named entity recognition, document classification, and semantic textual similarity (some datasets are borrowed from JMMLU and JMED-LLM). A tool called med-eval is developed to facilitate evaluation on JMedBench.	NII Aizawa Lab
Japanese Medical Language Model Evaluation Harness	A benchmark for evaluating Japanese LLMs in the medical domain in both Japanese and English, executable by a single command.	Individual (Issey Sukeda)
YakugakuQA	A Japanese pharmaceutical domain evaluation dataset based on national pharmacist licensing exams. Tests factual pharmaceutical knowledge.	EQUES Inc.
NayoseQA	A Japanese pharmaceutical domain evaluation dataset for cross-lingual terminology normalization. Tests understanding of synonyms and technical terms.	EQUES Inc.
SogoCheck	A novel task designed to assess consistency reasoning between paired statements. A challenging reasoning task where even GPT-4o performs poorly.	EQUES Inc.
MedRECT	A benchmark for evaluating the ability to detect and correct medical errors in clinical records. It consists of three tasks: error detection, error sentence identification, and error correction. It includes a Japanese version (663 samples) and an English version (458 samples), with the Japanese version constructed based on the Japanese Medical Licensing Examination.	Preferred Networks
karakuri-bench	A dataset for measuring performance of Japanese LLMs in customer support.	KARAKURI

Benchmarks for measuring factuality and safety

	Description	Developer
JTruthfulQA	The Japanese version of the dataset for evaluating the factuality of LLMs TruthfulQA. It includes questions about superstitions and other beliefs held by some people that are not factual, as well as questions about Japan-specific knowledge, all collected from scratch.	Waseda University Kawahara Lab
JCommonsenseMorality	A dataset on Japanese commonsense morality. Sentences describing actions are labeled with binary values indicating whether they are morally wrong or acceptable.	Hokkaido University Language Media Lab
JBBQ	The Japanese version of the social bias QA dataset BBQ, developed through translation, revision, and addition of questions based on Japanese culture and customs.	University of Tokyo Yanaka Lab

Benchmarks for measuring logical reasoning capabilities

	Description	Developer
JFLD (Japanese Formal Logic Deduction)	A dataset for evaluating deductive reasoning capabilities of Japanese LLMs (the Japanese version of the FLD (Formal Logic Deduction) proposed by the same authors). It is characterized by being composed of counterfactual samples to evaluate apart from the knowledge the LLM possesses.	Hitachi
JHumanEval	A Japanese version of the HumanEval benchmark, which assesses the ability to generate Python code from English instructions. In creating the Japanese version, the text was first machine-translated and then manually corrected.	Japan Women's University Kuramitsu Lab
JMultiPL-E	A dataset for evaluating code generation capabilities across 17 programming languages (C++, C#, Go, Java, JavaScript, PHP, Ruby, Rust, Scala, Swift, TypeScript, etc.) based on OpenAI HumanEval. Measures multilingual code understanding and generation performance.	Tohoku University Natural Language Processing Group

Benchmarks on instruction-following ability

	Description	Developer
LCTG Bench	A benchmark for the controllability of Japanese LLMs. It evaluates whether LLMs can adhere to constraints in four aspects: output format, character count, keywords, and forbidden words. The quality of the generated text is also evaluated.	CyberAgent
JFBench	A benchmark for evaluating the instruction-following ability of Japanese LLMs. In addition to 6 constraint groups translated from IFBench, 10 new groups specific to Japanese were created (e.g., polite/plain style, mixed hiragana/katakana/kanji, numerical notation). It has 16 constraint groups and 174 constraint types, evaluating a total of 1,600 samples with constraint counts of 1/2/4/8.	Preferred Networks

Benchmarks for embedding models

	Description	Developer
JMTEB	A benchmark developed as the Japanese version of MTEB. It consists of tasks such as document clustering, text classification, sentence similarity, sentence pair labeling prediction, and text extraction (a reranking task was recently added).	SB Intuitions
JQaRA	A dataset for evaluating Japanese document extraction and reranking accuracy. Each of the 1,667 questions is assigned 100 candidate documents, of which at least one can answer the question. The questions are taken from JAQKET, and the candidate documents are sourced from Japanese Wikipedia.	Individual (Yuichi Tateno)
JaCWIR	A dataset created for evaluating document extraction and reranking in domains other than Wikipedia. Each of the 5,000 questions is assigned one Web page that serves as the source of the question and 99 unrelated Web pages.	Individual (Yuichi Tateno)

Benchmarks for vision-language models

	Description	Developer
llm-jp-eval-mm	A framework for evaluating Japanese VLMs. It supports major Japanese tasks including Japanese-Heron-Bench, JA-VLM-Bench-In-the-Wild, JA-Multi-Image-VQA, JDocQA, JMMMU, JIC-VQA, MECHA-ja, CC-OCR, and CVQA, and provides LLM-as-a-Judge automatic scoring, fast batch inference via vLLM, and a web dashboard for inspecting results.	Research and Development Center for Large Language Models
simple-evals-mm	A framework for evaluating Japanese VLMs on 20 tasks (10 Japanese tasks + 10 English tasks). Japanese tasks include JAMMEval, BusinessSlideVQA, JMMMU, and MECHA-ja, while English tasks include AI2D, ChartQA, MMMU, and others.	Research and Development Center for Large Language Models
JAMMEval	A collection of evaluation datasets for Japanese VLMs, built by refining seven existing Japanese multimodal benchmarks (CC-OCR-JA, CVQA-JA, Heron-Bench, JA-Multi-Image-VQA, JA-VLM-Bench, JDocQA, JGraphQA).	Research and Development Center for Large Language Models
BusinessSlideVQA	A question-answer dataset with 220 questions about complex Japanese business slide images, designed to evaluate document comprehension capabilities.	Stockmark
JA-Business-Doc-RQ-Bench	A benchmark for evaluating multi-step reasoning ability on Japanese business documents. Consists of 229 questions (Yes/No, Factoid, and Numerical) requiring 3-5 reasoning steps over 4 image types (Chart, Table, Diagram, and Document). All images are synthetic, with questions and answers manually written.	Stockmark
JMMMU	A benchmark constructed as the Japanese version of MMMU Benchmark. It consists of 720 translated MMMU problems and 600 new problems unique to Japanese culture.	University of Tokyo Aizawa Lab
JDocQA	A question-answer dataset based on Japanese documents (pamphlets, slides, reports, websites), consisting of a total of 11,600 questions. It includes various question formats, including unanswerable questions.	NAIST Watanabe Lab
JGraphQA	A benchmark for evaluating Japanese chart comprehension. It consists of 100 images of four chart types (pie charts, line charts, bar charts, and tables) collected from Japanese corporate IR materials, with 2 manually created and verified question-answer pairs per image, for a total of 200 questions.	Ricoh
Heron VLM Leaderboard powered by Nejumi/WandB	Summarizes the evaluation results of Japanese-Heron-Bench and LLaVA-Bench-In-the-Wild (Japanese).	Turing, Weights & Biases
Japanese-Heron-Bench	21 images are assigned a total of 102 questions. It is characterized by image-question pairs that require knowledge related to Japan.	Turing
JA-VLM-Bench-In-the-Wild	A dataset independently prepared by Sakana AI to evaluate EvoVLM-JP-v1-7B. It consists of 50 questions assigned to 42 images. It is characterized by images and questions that require knowledge about Japan.	Sakana AI
JA-Multi-Image-VQA	A dataset for evaluating the question-answering ability in Japanese for multiple images.	Sakana AI
LLaVA-Bench-In-the-Wild (Japanese)	This is the Japanese version of LLaVA-Bench-In-the-Wild, translated using DeepL. It consists of 60 questions assigned to 24 images.	Turing
LLaVA-Bench (COCO) Japanese	This is the Japanese version, translated by DeepL, of the LLaVA-Bench (COCO) dataset used to evaluate LLaVA. It consists of 30 images, each with 3 types of questions assigned to them.	Turing
Japanese Visual Genome VQA dataset	A question-and-answer dataset annotated based on images from the Visual Genome dataset. A subset of this dataset, JA-VG-VQA-500, consisting of 500 questions, is sometimes used as a benchmark for evaluating VLMs.	Yahoo
japanese-bizform-table-kie	A benchmark for evaluating the accuracy of key information extraction from non-standard business forms. It consists of 50 form types with a total of 2,500 document images.	AI inside

Benchmarks/Datasets for Speech-Language Models

	Description	Developer
VoiceBench-ja	A benchmark that evaluates Japanese speech-language models (audio-input LLMs) by measuring the performance gap between audio input and text input. It consists of four subsets: Elyza (36 questions derived from ELYZA-tasks-100), Spoken-Elyza (34 questions refined for spoken dialogue), M-IFEval (172 instruction-following questions), and JamC-QA (1,452 multiple-choice questions on Japan-specific knowledge). The audio was synthesized with SB Intuitions' in-house TTS using speaker prompts from the JVS corpus (text is CC BY-SA 4.0; audio is not for commercial use or redistribution).	SB Intuitions
Joyo Kanji Yomi Benchmark	A benchmark for evaluating the kanji-level reading accuracy of Japanese TTS. It consists of 13,095 sentences covering all 2,136 Joyo kanji and their 4,378 readings, with each sentence written so that the context uniquely determines the target reading. Every sentence carries a full katakana reading annotation in which the substring corresponding to the target kanji is delimited, and all sentences were verified by 35 native Japanese speakers through a three-stage review. An evaluation toolkit that transcribes synthesized speech into katakana with Kana-Whisper and compares it against the reference reading is also available.	SB Intuitions

References for Models and Architectures

References for Training Methods

Our Contributors

We love contributors! Feel free to contribute to this project.

Citation

The summary of this repository is also published as a preprint: Exploring Open Large Language Models for the Japanese Language: A Practical Guide

When referencing this repository, please cite as follows:

@article{awesomeJapanese2024,
    title={{Exploring Open Large Language Models for the Japanese Language: A Practical Guide}},
    author={Kaito Sugimoto},
    doi={10.51094/jxiv.682},
    journal={Jxiv preprint},
    year={2024}
}

Footnotes

1. Some architectural changes have been made. For details, refer to: 1,000億パラメータ規模の独自LLM「PLaMo-100B」の事前学習 ↩

2. Refer to the following articles: 大規模言語モデルTanuki-8B, 8x8Bの位置づけや開発指針など, 大規模言語モデルを開発するにあたっての事前・事後学習の戦略メモー特に合成データについてー ↩ ↩²

3. Despite the license, the model card describes the model as a research-purpose release and states that commercial or mission-critical use is not intended. ↩

4. Some performance enhancements have been made to the original Llama model. See here for details. ↩

5. Details have not been made public but the private dataset includes data from the EleutherAI Polyglot project's Japanese team and from members of Stable Community Japan. ↩

6. This project conducted evaluation research on using right-to-left generation instead of the usual left-to-right generation, releasing both left-to-right and right-to-left models. ↩

7. Despite the license, the model card states that use is intended only for research and development purposes; direct use in actual clinical settings for disease diagnosis or clinical decision-making support is not recommended. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

8. Distributed for research purposes only; redistribution, commercial use, and direct use in actual clinical settings are not permitted. The model is also still at the base (pre-training) stage and has known limitations such as unstable end-of-sequence (EOD / EOS) token output and repetition. ↩

9. Before conducting Instruction Tuning, a Chat Vector between Llama 3 Instruct and Llama 3 Base is added. ↩ ↩²

10. After conducting Instruction Tuning, a Chat Vector between Llama 3 Instruct and Llama 3 Base is added. ↩ ↩²

11. However, if commercial use of KARAKURI LM is desired, direct contact with the developer, KARAKURI Inc., is required. ↩

12. In Instruction Tuning, because it uses data generated by OpenAI's models, such as GPT-3.5 and GPT-4, for training, there is a possibility that it may violate OpenAI's terms. ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰

13. Before conducting Instruction Tuning, a Chat Vector between Gemma 2 Instruct and Gemma 2 Base is added. ↩

14. Despite the license, the model card does not recommend direct use in actual clinical settings for disease diagnosis or clinical decision-making support, and recommends limiting use to an information-providing tool that assists the judgment of medical professionals. ↩ ↩² ↩³

15. Although the base model is not officially disclosed, the config.json architecture is DeepseekV3ForCausalLM, the tokenizer is identical to DeepSeek-V3, and the repository includes a DeepSeek NOTICE file, strongly suggesting it is based on DeepSeek-V3. ↩

16. ○: The model is on the HuggingFace Model Hub and can be loaded in with the AutoModel.from_pretrained() command. △: The model is not on the Model Hub but can be loaded in manually with the HuggingFace transformers library. ✕: The model is not directly loadable with HuggingFace. ↩

17. By removing Causal Attention from Llama, it is used as an encoder-type model. ↩ ↩²

18. This project conducted evaluation research on pre-tokenization morphological analysis and released their best performing model, which used Juman++ and BPE. ↩

19. However, the maximum sequence length has been extended to 2048, and various architectural changes have been made compared to the original BERT. See the HuggingFace repository README for details. ↩

20. nlp-waseda/roberta-base-japanese and nlp-waseda/roberta-large-japanese trained using a 128 token context length, but nlp-waseda/roberta-large-japanese-seq512 expanded the context length to 512. ↩

21. Extended to a 1282 context length from the usual 512. ↩

22. For details of each model, please refer to Chapter 4 of the authors' paper. Note that the SC-2M-wiki model is strictly not a domain-specific model as it is pre-trained only on Wikipedia. ↩

23. The "small" model trains on Japanese Wikipedia and the Japanese Financial Corpus simultaneously, while the "base" model takes the TohokuUniversityBERT and conducts additional training on the Japanese Financial Corpus. ↩

24. ManbyoWordPiece conducts a pre-tokenization step using MeCab (IPA+Manbyo dictionaries) and uses WordPiece for subword tokenization, while the SentencePiece model tokenizes text directly using a unigram model. ↩

25. The classification of embedding models was referenced from Dense Text Retrieval based on Pretrained Language Models: A Survey (Zhao+, 2022). The Bi-Encoder architecture inputs two separate inputs into the model and vectorizes each, using their dot product or cosine similarity as a measure of their proximity. In contrast, the Cross-Encoder architecture inputs the combined inputs into the model to directly compute their proximity internally. Although Cross-Encoders incur higher computational costs, they are often used as rerankers in information extraction due to their ability to compute input proximity more precisely. Among Bi-Encoders, there are types (e.g., ColBERT) that represent the input as multiple vectors (such as one per token) rather than a single vector, hence further classification into Single-representation bi-encoders and Multi-representation bi-encoders. ↩

26. However, it calls for consideration for use in research and education. Additionally, be aware that some of the licenses for the source models are not Apache 2.0. ↩ ↩² ↩³

27. This model transcribes speech into katakana sequences rather than ordinary Japanese orthography. It was developed as the backbone of Kana-CER, a metric for measuring the kanji reading accuracy of TTS systems. ↩