Open models you can run on-prem

Architecture numbers from each model's published configuration, memory at common precisions, and the hardware that runs it. Open any model to size it for your users, or build a full architecture.

ModelVendorTypeParamsContextFP8 weights4-bit weightsLicence
Qwen3-VL-8B-InstructAlibaba QwenDense8.8B256K8.80 GB4.93 GBApache-2.0
Qwen3.5-9BAlibaba QwenDense9.7B256K9.70 GB5.43 GBApache-2.0
Qwen3.8-27BAlibaba QwenDense27.8B256K27.8 GB15.6 GBApache-2.0
Qwen3-30B-A3B-Instruct-2507Alibaba QwenMoE30.5B / 3.3B active256K30.5 GB17.1 GBApache-2.0
Qwen3-32BAlibaba QwenDense32.8B128K32.8 GB18.4 GBApache-2.0
Qwen3.6-35B-A3BAlibaba QwenMoE36B / 3B active256K36.0 GB20.2 GBApache-2.0
Qwen3-Coder-NextAlibaba QwenMoE79.7B / 3B active256K79.7 GB44.6 GBApache-2.0
Qwen3.5-122B-A10BAlibaba QwenMoE125B / 10B active256K125 GB70.0 GBApache-2.0
Qwen3.8-Flash-NextAlibaba QwenMoE180B / 6B active256K180 GB101 GBQwen Community License 1.0
Qwen3.5-397B-A17BAlibaba QwenMoE403B / 17B active256K403 GB226 GBApache-2.0
Qwen3.8-2.4T-A95BAlibaba QwenMoE2446B / 95B active256K2446 GB1370 GBQwen3.8-Max License
Olmo 3.1 32B ThinkAllen Institute for AI (Ai2)Dense32B64K32.0 GB17.9 GBApache-2.0
Ling-3.0-flashAnt Group inclusionAIMoE124B / 5.1B active256K124 GB69.4 GBMIT
Trinity Large ThinkingArcee AIMoE398B / 13B active512K398 GB223 GBOpenMDW-1.1
Command A (03-2025)CohereDense111B256K111 GB62.2 GBCC-BY-NC-4.0
Command A+ (05-2026)CohereMoE218B / 25B active128K218 GB122 GBApache-2.0
DeepSeek-V3.2DeepSeekMoE671B / 37B active160K671 GB376 GBMIT
DeepSeek-V4.1-Flash-1017DeepSeekMoE748B / 16B active1024K748 GB419 GBMIT
DeepSeek-V4-Pro-0813DeepSeekMoE1600B / 49B active1024K1600 GB896 GBMIT
Gemma 4 12B ITGoogleDense11.95B256K11.9 GB6.69 GBApache-2.0
Gemma 4 26B A4B ITGoogleMoE25.2B / 3.8B active256K25.2 GB14.1 GBApache-2.0
Gemma 3 27B ITGoogleDense27B128K27.0 GB15.1 GBGemma Terms of Use
Gemma 4 31B ITGoogleDense30.7B256K30.7 GB17.2 GBApache-2.0
Granite 4.2 8BIBMDense8B128K8.00 GB4.48 GBApache-2.0
Granite 4.2 30BIBMDense30B128K30.0 GB16.8 GBApache-2.0
Llama 3.1 8B InstructMetaDense8B128K8.00 GB4.48 GBLlama 3.1 Community License
Llama 3.3 70B InstructMetaDense70B128K70.0 GB39.2 GBLlama 3.3 Community License
Llama 4 Scout 17B-16E InstructMetaMoE109B / 17B active10240K109 GB61.0 GBLlama 4 Community License
Llama 4 Maverick 17B-128E InstructMetaMoE400B / 17B active1024K400 GB224 GBLlama 4 Community License
Muse Spark Open 816B-A42BMetaMoE816B / 42B active1024K816 GB457 GBMuse Model License
Phi-4MicrosoftDense14B16K14.0 GB7.84 GBMIT
MiniMax-M3MiniMaxMoE428B / 23B active1024K428 GB240 GBMiniMax Community License
Mistral Small 3.2 24B Instruct 2506Mistral AIDense24B128K24.0 GB13.4 GBApache-2.0
Magistral Small 1.2 (2509)Mistral AIDense24B128K24.0 GB13.4 GBApache-2.0
Devstral Small 2 24B Instruct 2512Mistral AIDense24B256K24.0 GB13.4 GBApache-2.0
Mistral Small 4 119B A6B (2603)Mistral AIMoE119B / 6.5B active256K119 GB66.6 GBApache-2.0
Mistral Medium 3.5 128BMistral AIDense128B256K128 GB71.7 GBModified MIT License
Mistral Large 3 675B Instruct 2512Mistral AIMoE675B / 41B active256K675 GB378 GBApache-2.0
Kimi-K2.6Moonshot AIMoE1000B / 32B active256K1000 GB560 GBModified MIT
Kimi-K2.7-CodeMoonshot AIMoE1000B / 32B active256K1000 GB560 GBModified MIT
Kimi-K3Moonshot AIMoE2800B / 104B active1024K2800 GB1568 GBKimi K3 License
NVIDIA Nemotron 3.5 Lightning 30B-A3BNVIDIAMoE30B / 3B active1024K30.0 GB16.8 GBOpenMDW-1.1
NVIDIA Nemotron 3 Super 120B-A12BNVIDIAMoE120B / 12B active1024K120 GB67.2 GBNVIDIA Nemotron Open Model License
gpt-oss-20bOpenAIMoE21B / 3.6B active128K21.0 GB11.8 GBApache-2.0
gpt-oss-120bOpenAIMoE117B / 5.1B active128K117 GB65.5 GBApache-2.0
Step-3.7-FlashStepFunMoE198B / 11B active256K198 GB111 GBApache-2.0
Hy3Tencent HunyuanMoE295B / 21B active256K295 GB165 GBApache-2.0
MiMo-V2.6-Flash-RLXiaomi MiMoMoE309B / 15B active1024K309 GB173 GBMIT
MiMo-V2.6-Pro-RLXiaomi MiMoMoE1020B / 42B active1024K1020 GB571 GBMIT
GLM-4.7-FlashZhipu AI (Z.ai)MoE31.2B / 3B active198K31.2 GB17.5 GBMIT
GLM-5.3-FlashZhipu AI (Z.ai)MoE320B / 18B active1024K320 GB179 GBMIT
GLM-5.3Zhipu AI (Z.ai)MoE744B / 40B active1024K744 GB417 GBGLM-5.3 License