Open source LLM leaderboard
120 open-weight language models from 24 developers, 107 of them ranked on LMArena’s text leaderboard, in its order. What open weights are and what the licences allow is explained in open-weight AI models explained.
| LMArena rank | Coding rank | Model | Developer | Licence | Context | Released |
|---|---|---|---|---|---|---|
| #18 | #11 | Kimi K3 | Moonshot AI | Open weights (Kimi K3 License) | 1M tokens | Jul 2026 |
| #23 | #29 | Qwen3.8 | Alibaba | Open weights under a licence named "qwen3.8-max" - NOT Apache 2.0 (the Qwen3.8-2 | 1M tokens (hosted Qwen3.8-Max); open weights 262k native, extensible to ~1.01M | Aug 2026 |
| #26 | #33 | GLM-5.3 | Zhipu AI | GLM-5.3 License (custom: MIT-style terms plus a security-review clause for MaaS | 1M tokens | Aug 2026 |
| #28 | #12 | MiMo V2.6 Pro | Xiaomi | MIT | 1M tokens | Sept 2026 |
| #34 | #30 | DeepSeek V4.1-Flash | DeepSeek | MIT | 1M tokens | Sept 2026 |
| #35 | #27 | GLM-5.3-Flash | Zhipu AI | MIT (weights on Hugging Face, zai-org/GLM-5.3-Flash) | 1M tokens | Aug 2026 |
| #37 | #46 | GLM-5.2 | Zhipu AI | Open weights (MIT) | 1M tokens | Jun 2026 |
| #51 | #35 | MiMo-V2.5-Pro | Xiaomi | MIT | 1M tokens | Apr 2026 |
| #53 | #51 | GLM-5.1 | Zhipu AI | Open weights (MIT) | 202.8K tokens | Apr 2026 |
| #59 | #44 | Kimi K2.6 | Moonshot AI | Open weights (Modified MIT) | 262.1K tokens | Apr 2026 |
| #64 | #79 | DeepSeek V4-Pro | DeepSeek | MIT | 1M tokens | Aug 2026 |
| #65 | #85 | GLM 5 | Zhipu AI | MIT | 202.8K tokens | — |
| #74 | #73 | HY3 | Tencent | Apache 2.0 | 262.1K tokens | — |
| #76 | #74 | Gemma 4 31B | Apache 2.0 | 262.1K tokens | — | |
| #78 | #75 | Kimi K2.5 | Moonshot AI | Modified MIT | 256K tokens | — |
| #81 | #57 | MiMo V2.6 Flash | Xiaomi | MIT | 1M tokens | Sept 2026 |
| #92 | #94 | Qwen3.5 397B A17b | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #94 | #90 | Inkling | Thinking Machines | Apache 2.0 | 256K tokens | — |
| #95 | #107 | GLM 4.7 | Zhipu AI | MIT | 204.8K tokens | — |
| #97 | #88 | MiniMax-M3 | MiniMax | Open weights (MiniMax Community License) | 1M tokens (512k guaranteed minimum) | Jun 2026 |
| #98 | #114 | DeepSeek V4 Flash High Preview | DeepSeek | MIT | 1M tokens | — |
| #100 | #108 | Gemma 4 26B A4b | Apache 2.0 | 256K tokens | — | |
| #102 | #82 | Qwen3.8 27B | Alibaba | Apache 2.0 | 1M tokens | Aug 2026 |
| #105 | #106 | DeepSeek V4-Flash | DeepSeek | MIT | 1M tokens | Jul 2026 |
| #109 | #95 | MiMo V2.5 | Xiaomi | MIT | 1M tokens | — |
| #117 | #104 | Kimi K2 Thinking Turbo | Moonshot AI | Modified MIT | 262.1K tokens | — |
| #119 | #116 | Mistral Medium 3.5 | Mistral AI | Open weights (Modified MIT) | 256k tokens | Apr 2026 |
| #122 | #121 | Nemotron 3 Ultra | NVIDIA | Open weights (OpenMDW-1.1) | 1M tokens | Jun 2026 |
| #125 | #115 | Muse Glimmer | Meta | Apache-2.0 | 131.1K tokens | Aug 2026 |
| #127 | #145 | GLM 4.6 | Zhipu AI | MIT | 204.8K tokens | — |
| #130 | #124 | Qwen3 235B A22b Instruct 2507 | Alibaba | Apache 2.0 | — | — |
| #133 | #130 | Kimi K2 0905 Preview | Moonshot AI | Modified MIT | 262.1K tokens | — |
| #136 | #142 | Kimi K2 0711 Preview | Moonshot AI | Modified MIT | 131.1K tokens | — |
| #138 | #144 | Qwen3.5 122B A10b | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #141 | #113 | MiniMax M2.7 | MiniMax | Modified MIT | 204.8K tokens | Mar 2026 |
| #145 | #126 | Mistral Large 3 | Mistral AI | Open weights (Apache 2.0) | 256k tokens | Dec 2025 |
| #147 | #136 | Qwen3 VL 235B A22b Instruct | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #149 | #149 | Nemotron 3.5 Super VL | NVIDIA | OpenMDW License Agreement 1.1 | 1M tokens | Oct 2026 |
| #151 | #153 | GLM 4.5 | Zhipu AI | MIT | 131.1K tokens | — |
| #154 | #155 | Hunyuan HY3 Preview | Tencent | tencent-hunyuan-community | 262.1K tokens | — |
| #155 | #158 | Qwen3.5 27B | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #158 | #128 | Inkling Small | Thinking Machines | Apache 2.0 | 256K tokens | — |
| #161 | #165 | Qwen3 235B A22b No Thinking | Alibaba | Apache 2.0 | 131.1K tokens | — |
| #165 | #169 | Qwen3 235B A22b Thinking 2507 | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #166 | #164 | Qwen3 Next 80B A3b Instruct | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #167 | #166 | DeepSeek R1 | DeepSeek | MIT | 128K tokens | — |
| #170 | #157 | Qwen3 VL 235B A22b Thinking | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #172 | #178 | Qwen3.5 35B A3b | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #174 | #159 | Step 3.5 Flash | StepFun | Apache 2.0 | 262.1K tokens | — |
| #178 | #163 | MiniMax M2.5 | MiniMax | Modified MIT | 204.8K tokens | — |
| #182 | #150 | Qwen3 Coder 480B A35b Instruct | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #186 | #184 | MiMo V2 Flash | Xiaomi | MIT | 262.1K tokens | — |
| #187 | #172 | MiniMax M2.1 Preview | MiniMax | Modified MIT | 204.8K tokens | Dec 2025 |
| #190 | #173 | Qwen3 30B A3b Instruct 2507 | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #193 | #201 | GLM 4.6V | Zhipu AI | MIT | 131.1K tokens | — |
| #198 | #193 | GLM 4.5 Air | Zhipu AI | MIT | 131.1K tokens | — |
| #200 | #196 | Qwen3 Next 80B A3b Thinking | Alibaba | Apache 2.0 | 262.1K tokens | — |
| #202 | #264 | Gemma 3 27B IT | Gemma | 128K tokens | — | |
| #205 | #199 | MiniMax M1 | MiniMax | Apache 2.0 | 1M tokens | — |
| #208 | #208 | Nemotron 3 Super | NVIDIA | Open weights (NVIDIA Nemotron Open Model License) | 1M tokens (256k default, VRAM-limited) | Mar 2026 |
| #212 | #204 | Mistral Small 2506 | Mistral AI | Apache 2.0 | 128K tokens | Jun 2025 |
| #214 | #226 | Command A 03 2025 | Cohere | CC-BY-NC-4.0 | 256K tokens | — |
| #216 | #192 | Nemotron 3.5 Lightning | NVIDIA | OpenMDW License Agreement 1.1 (approved for commercial use) | 1M tokens (256k in NVIDIA’s single-H100 config) | Aug 2026 |
| #217 | #211 | GLM 4.5V | Zhipu AI | MIT | 65.5K tokens | — |
| #218 | #225 | gpt-oss-120b | OpenAI | Apache 2.0 | 131k tokens | Aug 2025 |
| #220 | #207 | Step 3 | StepFun | Apache 2.0 | 65.5K tokens | — |
| #224 | #223 | Llama 3.1 Nemotron Ultra 253B V1 | NVIDIA | Nvidia Open Model | 131.1K tokens | Apr 2025 |
| #225 | #209 | Qwen3 32B | Alibaba | Apache 2.0 | 131.1K tokens | — |
| #229 | #235 | MiniMax M2 | MiniMax | Modified MIT | 204.8K tokens | Oct 2025 |
| #230 | #206 | Ling Flash 2 | Ant Group | MIT | 128K tokens (with YaRN) | Sept 2025 |
| #232 | #210 | NVIDIA Llama 3.3 Nemotron Super 49B V1.5 | NVIDIA | Nvidia Open | 131.1K tokens | Jul 2025 |
| #235 | #297 | Gemma 3 12B IT | Gemma | 131.1K tokens | — | |
| #242 | #234 | QwQ 32B | Alibaba | Apache 2.0 | 131.1K tokens | — |
| #244 | #240 | Llama 3.1 405B Instruct Bf16 | Meta | Llama 3.1 Community | 128K tokens | Jul 2024 |
| #246 | #247 | Llama 3.1 405B Instruct Fp8 | Meta | Llama 3.1 Community | 128K tokens | Jul 2024 |
| #248 | #237 | OLMo 3.1 32B Instruct | Ai2 | Apache 2.0 | 65.5K tokens | — |
| #250 | #257 | Llama 3.3 Nemotron 49B Super V1 | NVIDIA | NVIDIA Open Model License | 128K tokens | Mar 2025 |
| #251 | #243 | Llama 4 Maverick 17B 128E Instruct | Meta | Llama 4 | 1M tokens | Apr 2025 |
| #252 | #233 | Qwen3 30B A3b | Alibaba | Apache 2.0 | 131.1K tokens | — |
| #258 | #224 | Ring Flash 2 | Ant Group | MIT | — | Sept 2025 |
| #262 | #256 | Llama 4 Scout | Meta | Open weights (Llama 4 Community License Agreement) | 10M tokens | Apr 2025 |
| #265 | #271 | Qwen Max 0919 | Alibaba | Qwen | 32.8K tokens | — |
| #266 | #279 | Llama 3.3 70B Instruct | Meta | Llama-3.3 | 131.1K tokens | Dec 2024 |
| #268 | #307 | Gemma 3N E4b IT | Gemma | 32.8K tokens | — | |
| #269 | #251 | gpt-oss-20b | OpenAI | Apache 2.0 | 131.1K tokens | — |
| #271 | #269 | Mistral Large 2407 | Mistral AI | Mistral Research | 131.1K tokens | — |
| #273 | #258 | NVIDIA Nemotron 3 Nano 30B A3b Bf16 | NVIDIA | NVIDIA Nemotron Open Model License | 1M tokens (256K default) | Dec 2025 |
| #281 | #260 | OLMo 3 32B Think | Ai2 | Apache 2.0 | 65.5K tokens | — |
| #283 | #278 | Mistral Large 2411 | Mistral AI | MRL | 128K tokens | — |
| #285 | #270 | Granite 4.1 8B | IBM | Apache 2.0 | 131.1K tokens | Apr 2026 |
| #286 | #329 | Gemma 3 4B IT | Gemma | 131.1K tokens | — | |
| #287 | #268 | Qwen2.5 72B Instruct | Alibaba | Qwen | — | — |
| #288 | #259 | Mistral Small 3.1 24B Instruct 2503 | Mistral AI | Apache 2.0 | 128K tokens | — |
| #289 | #294 | Llama 3.1 Nemotron 70B Instruct | NVIDIA | NVIDIA Open Model License (plus Llama 3.1 Community License) | 131.1K tokens | Oct 2024 |
| #291 | #288 | Llama 3.1 70B Instruct | Meta | Llama 3.1 Community | 131.1K tokens | Jul 2024 |
| #295 | #311 | Gemma 2 27B IT | Gemma license | 8.2K tokens | — | |
| #300 | #300 | Llama 3.1 Nemotron 51B Instruct | NVIDIA | NVIDIA Open Model License | — | Sept 2024 |
| #303 | #306 | Llama 3.1 Tulu 3 70B | Ai2 | Llama 3.1 Community License Agreement | — | Nov 2024 |
| #306 | #305 | Nemotron 4 340B Instruct | NVIDIA | NVIDIA Open Model License | 4,096 tokens | Jun 2024 |
| #308 | #320 | Command R Plus 08 2024 | Cohere | CC-BY-NC-4.0 | 128K tokens | — |
| #314 | #316 | C4AI Aya Expanse 32B | Cohere | CC-BY-NC-4.0 (non-commercial) | 128K tokens | Oct 2024 |
| #322 | #309 | Phi 4 | Microsoft AI | MIT | 16.4K tokens | Dec 2024 |
| #323 | #323 | OLMo 2 0325 32B Instruct | Ai2 | Apache-2.0 | 4.1K tokens | — |
| #328 | #328 | Ministral 8B 2410 | Mistral AI | MRL | 131.1K tokens | — |
| #340 | #340 | C4AI Aya Expanse 8B | Cohere | CC-BY-NC-4.0 (non-commercial) | 8K tokens | Oct 2024 |
| #371 | #377 | Llama 3.2 3B Instruct | Meta | Llama 3.2 | 131.1K tokens | Sept 2024 |
| #400 | #392 | Llama 3.2 1B Instruct | Meta | Llama 3.2 | 128K tokens | — |
Open-weight models not yet ranked
LMArena has no text-board row for these yet; newest first.
- Spark X2.6 · iFLYTEK · Oct 2026
- Muse Spark Open · Meta · Oct 2026
- Nemotron 3.5 Bolt 74B-A9B · NVIDIA · Oct 2026
- Beam 1.0 · Reflection AI · Oct 2026
- Hy4 preview · Tencent · Aug 2026
- Qwen3.8-Flash-Next · Alibaba · Aug 2026
- K-EXAONE 2.0 · LG AI Research · Jul 2026
- Kimi K2.7 Code · Moonshot AI · Jun 2026
- Step 3.7 Flash · StepFun · May 2026
- Command A+ · Cohere · May 2026
- Laguna M.1 · Poolside · Apr 2026
- Laguna XS.2 · Poolside · Apr 2026
- Falcon H1 34B Instruct · Technology Innovation Institute · May 2025
Frequently asked questions
Which open-source LLM ranks highest?+
On LMArena’s text leaderboard the highest-ranked open-weight model listed here is Kimi K3 from Moonshot AI, at rank 18 overall. The full table, open and closed models together, is at aiblogly.com/benchmarks.
Which open-source LLM ranks highest for coding?+
In LMArena’s coding category, Kimi K3 from Moonshot AI is the highest open-weight model listed here, at rank 11.
What counts as an open-source LLM here?+
A model whose developer publishes its weights for download, under any licence - Apache 2.0, MIT, or a custom licence such as Llama’s or Gemma’s. Several of those licences limit commercial use; each model’s licence is in the table and on its page.
Related: Chinese AI models · running open models locally · free AI API keys