Headroom

Models

138 open models from 18 makers, with their published architecture. Pick one to see memory per format, KV cache per user and which GPUs it fits on.

138 models
DeepSeek V4.1 FlashDeepSeek · DeepSeek V4763B · 16.0B act.1,024KMQA1,422 GB713 GB381 GB2026-09
Qwen3.8 27BAlibaba Qwen · Qwen3.827.8B256Khybrid51.7 GB30.6 GB20.7 GB2026-08
Qwen3.8 2.4T-A95BAlibaba Qwen · Qwen3.82,446B · 95.3B act.256Khybrid4,556 GB2,286 GB1,221 GB2026-08
GLM-5.3Zhipu AI (Z.ai) · GLM-5753B · 40.0B act.1,024KMLA1,403 GB705 GB378 GB2026-08
GLM-5.3-FlashZhipu AI (Z.ai) · GLM-5321B · 18.0B act.1,024Khybrid599 GB302 GB162 GB2026-08
Nemotron 3.5 Lightning 30B-A3BNVIDIA · Nemotron 331.6B · 3.00B act.1,024Khybrid58.8 GB30.7 GB17.6 GB2026-08
Qwen3.8-Flash-Next 180B-A6BAlibaba Qwen · Qwen3.8180B · 6.00B act.256Khybrid335 GB170 GB92.5 GB2026-08
Kimi K3Moonshot AI · Kimi K32,780B · 104B act.1,024Khybrid5,178 GB2,593 GB1,382 GB2026-07
Gemma 4 12BGoogle · Gemma 412.0B256KGQA22.3 GB12.1 GB7.3 GB2026-06
GLM-5.2Zhipu AI (Z.ai) · GLM-5753B · 40.0B act.1,024KMLA1,403 GB705 GB378 GB2026-06
MiniMax-M3MiniMax · MiniMax427B · 23.0B act.1,024KGQA795 GB402 GB218 GB2026-06
Nemotron 3 Ultra 550B-A55BNVIDIA · Nemotron 3561B · 55.0B act.1,024Khybrid1,044 GB526 GB283 GB2026-06
Command A+ (05-2026)Cohere · Command219B · 25.0B act.128KGQA407 GB205 GB110 GB2026-05
Qwen3.6 35B-A3BAlibaba Qwen · Qwen3.636.0B · 3.45B act.256Khybrid67.0 GB35.4 GB20.6 GB2026-04
Qwen3.6 27BAlibaba Qwen · Qwen3.627.8B256Khybrid51.7 GB30.6 GB20.7 GB2026-04
DeepSeek V4 ProDeepSeek · DeepSeek V41,599B · 49.0B act.1,024KMQA2,978 GB1,492 GB796 GB2026-04
DeepSeek V4 FlashDeepSeek · DeepSeek V4291B · 13.0B act.1,024KMQA542 GB273 GB147 GB2026-04
Mistral Medium 3.5 128BMistral AI · Mistral Medium128B256KGQA238 GB125 GB72.0 GB2026-04
Kimi K2.6Moonshot AI · Kimi K21,027B · 32.9B act.256KMLA1,913 GB961 GB514 GB2026-04
MiniMax-M2.7MiniMax · MiniMax229B · 11.0B act.200KGQA426 GB215 GB117 GB2026-04
Granite 4.1 3BIBM · Granite 4.13.40B128KGQA6.3 GB3.4 GB2.0 GB2026-04
Granite 4.1 8BIBM · Granite 4.18.38B128KGQA15.6 GB8.2 GB4.7 GB2026-04
Granite 4.1 30BIBM · Granite 4.128.9B128KGQA53.8 GB27.3 GB14.8 GB2026-04
Gemma 4 31BGoogle · Gemma 431.3B256KGQA58.3 GB30.4 GB17.4 GB2026-03
Gemma 4 26B A4BGoogle · Gemma 425.8B · 3.82B act.256KGQA48.1 GB24.7 GB13.8 GB2026-03
Gemma 4 E4BGoogle · Gemma 48.00B128KGQA14.9 GB8.1 GB4.9 GB2026-03
Gemma 4 E2BGoogle · Gemma 45.12B128KMQA9.5 GB5.1 GB3.1 GB2026-03
Mistral Small 4 119BMistral AI · Mistral Small119B · 6.50B act.256KMLA222 GB113 GB62.0 GB2026-03
Nemotron 3 Nano 4BNVIDIA · Nemotron 33.97B256Khybrid7.4 GB5.2 GB4.2 GB2026-03
Nemotron 3 Super 120B-A12BNVIDIA · Nemotron 3124B · 12.0B act.1,024Khybrid230 GB117 GB64.1 GB2026-03
Qwen3-Coder-Next 80B-A3BAlibaba Qwen · Qwen3-Coder79.7B · 3.87B act.256Khybrid148 GB75.4 GB41.1 GB2026-02
Qwen3.5 397B-A17BAlibaba Qwen · Qwen3.5403B · 17.4B act.256Khybrid751 GB379 GB205 GB2026-02
Qwen3.5 122B-A10BAlibaba Qwen · Qwen3.5125B · 9.77B act.256Khybrid233 GB119 GB66.1 GB2026-02
Qwen3.5 35B-A3BAlibaba Qwen · Qwen3.536.0B · 3.45B act.256Khybrid67.0 GB35.4 GB20.6 GB2026-02
Qwen3.5 27BAlibaba Qwen · Qwen3.527.8B256Khybrid51.7 GB30.6 GB20.7 GB2026-02
Qwen3.5 9BAlibaba Qwen · Qwen3.59.65B256Khybrid18.0 GB12.8 GB10.3 GB2026-02
Qwen3.5 4BAlibaba Qwen · Qwen3.54.66B256Khybrid8.7 GB4.9 GB3.2 GB2026-02
Qwen3.5 2BAlibaba Qwen · Qwen3.52.27B256Khybrid4.2 GB2.6 GB1.8 GB2026-02
Qwen3.5 0.8BAlibaba Qwen · Qwen3.50.87B256Khybrid1.6 GB1.0 GB0.8 GB2026-02
GLM-5 / GLM-5.1Zhipu AI (Z.ai) · GLM-5754B · 40.0B act.198KMLA1,404 GB706 GB378 GB2026-02
Kimi K2.5Moonshot AI · Kimi K21,027B · 32.9B act.256KMLA1,913 GB961 GB514 GB2026-01
GLM-4.7-Flash 30B-A3BZhipu AI (Z.ai) · GLM-4.531.2B · 3.90B act.198KMLA58.2 GB30.3 GB17.2 GB2026-01
DeepSeek V3.2DeepSeek · DeepSeek V3685B · 37.6B act.160KMLA1,277 GB642 GB344 GB2025-12
Devstral Small 2 24BMistral AI · Devstral24.0B384KGQA44.7 GB24.9 GB15.6 GB2025-12
Ministral 3 3BMistral AI · Ministral 33.85B256KGQA7.2 GB4.0 GB2.5 GB2025-12
Ministral 3 8BMistral AI · Ministral 38.92B256KGQA16.6 GB10.3 GB7.4 GB2025-12
Ministral 3 14BMistral AI · Ministral 313.9B256KGQA26.0 GB15.5 GB10.6 GB2025-12
Mistral Large 3 (675B)Mistral AI · Mistral Large675B · 40.0B act.256KMLA1,257 GB632 GB339 GB2025-12
Devstral 2 123BMistral AI · Devstral125B256KGQA233 GB122 GB70.7 GB2025-12
GLM-4.7Zhipu AI (Z.ai) · GLM-4.5358B · 33.6B act.198KGQA667 GB337 GB182 GB2025-12
Nemotron 3 Nano 30B-A3BNVIDIA · Nemotron 331.6B · 3.50B act.1,024Khybrid58.8 GB30.7 GB17.6 GB2025-12
Olmo 3.1 32BAllenAI · Olmo 332.2B64KGQA60.0 GB31.9 GB18.8 GB2025-12
Kimi K2 ThinkingMoonshot AI · Kimi K21,026B · 32.9B act.256KMLA1,912 GB960 GB514 GB2025-11
Olmo 3 7BAllenAI · Olmo 37.30B64KMHA13.6 GB8.3 GB5.9 GB2025-11
Kimi Linear 48B-A3BMoonshot AI · Kimi Linear49.1B · 3.48B act.1,024Khybrid91.5 GB47.2 GB26.4 GB2025-10
MiniMax-M2 / M2.1 / M2.5MiniMax · MiniMax229B · 11.0B act.192KGQA426 GB215 GB117 GB2025-10
Granite 4.0 H-Small 32B-A9BIBM · Granite 4.032.2B · 8.80B act.128Khybrid60.0 GB30.4 GB16.5 GB2025-10
Granite 4.0 H-Tiny 7B-A1BIBM · Granite 4.06.94B · 1.47B act.128Khybrid12.9 GB6.6 GB3.6 GB2025-10
Granite 4.0 H-Micro 3BIBM · Granite 4.03.19B128Khybrid5.9 GB3.2 GB1.9 GB2025-10
Qwen3-Next 80B-A3BAlibaba Qwen · Qwen3-Next81.3B · 3.87B act.256Khybrid151 GB76.9 GB41.9 GB2025-09
Magistral Small 1.2 (2509)Mistral AI · Mistral Small24.0B128KGQA44.7 GB24.9 GB15.6 GB2025-09
Kimi K2 Instruct (0905)Moonshot AI · Kimi K21,026B · 32.9B act.256KMLA1,912 GB960 GB514 GB2025-09
GLM-4.6Zhipu AI (Z.ai) · GLM-4.5357B · 33.6B act.198KGQA665 GB335 GB181 GB2025-09
Gemma 3 270MGoogle · Gemma 30.27B32KMQA0.5 GB0.4 GB0.4 GB2025-08
DeepSeek V3.1DeepSeek · DeepSeek V3685B · 37.6B act.160KMLA1,275 GB641 GB344 GB2025-08
gpt-oss-120bOpenAI · gpt-oss117B · 5.13B act.128KGQA218 GB111 GB61.0 GB2025-08
gpt-oss-20bOpenAI · gpt-oss20.9B · 3.61B act.128KGQA38.9 GB21.6 GB13.5 GB2025-08
Nemotron Nano 9B v2NVIDIA · Nemotron8.89B128Khybrid16.6 GB10.5 GB7.6 GB2025-08
Nemotron Nano 12B v2NVIDIA · Nemotron12.3B128Khybrid22.9 GB14.0 GB9.8 GB2025-08
Qwen3-Coder 30B-A3BAlibaba Qwen · Qwen3-Coder30.5B · 3.35B act.256KGQA56.9 GB29.6 GB16.8 GB2025-07
Qwen3-Coder 480B-A35BAlibaba Qwen · Qwen3-Coder480B · 35.5B act.256KGQA894 GB451 GB243 GB2025-07
GLM-4.5Zhipu AI (Z.ai) · GLM-4.5358B · 33.6B act.128KGQA667 GB337 GB182 GB2025-07
GLM-4.5-AirZhipu AI (Z.ai) · GLM-4.5110B · 13.4B act.128KGQA206 GB105 GB58.1 GB2025-07
Llama-3.3-Nemotron-Super-49B v1.5NVIDIA · Nemotron49.9B128KGQA92.9 GB50.4 GB30.4 GB2025-07
SmolLM3 3BHugging Face · SmolLM3.08B128KGQA5.7 GB3.1 GB1.9 GB2025-07
Gemma 3n E4BGoogle · Gemma 3n7.85B32KGQA14.6 GB7.8 GB4.6 GB2025-06
Gemma 3n E2BGoogle · Gemma 3n5.44B32KGQA10.1 GB5.6 GB3.4 GB2025-06
Mistral Small 3.2 24BMistral AI · Mistral Small24.0B128KGQA44.7 GB24.9 GB15.6 GB2025-06
MiniMax-M1 (80k)MiniMax · MiniMax456B · 45.9B act.977Khybrid850 GB429 GB232 GB2025-06
DeepSeek R1-0528DeepSeek · DeepSeek R1685B · 37.6B act.160KMLA1,275 GB641 GB344 GB2025-05
Llama 4 Scout (17B-16E)Meta · Llama 4109B · 17.2B act.10,240KGQA202 GB105 GB59.4 GB2025-04
Llama 4 Maverick (17B-128E)Meta · Llama 4402B · 17.2B act.1,024KGQA748 GB378 GB204 GB2025-04
Qwen3 0.6BAlibaba Qwen · Qwen30.60B128KGQA1.1 GB0.7 GB0.5 GB2025-04
Qwen3 1.7BAlibaba Qwen · Qwen31.72B128KGQA3.2 GB1.9 GB1.3 GB2025-04
Qwen3 4BAlibaba Qwen · Qwen34.02B128KGQA7.5 GB4.1 GB2.5 GB2025-04
Qwen3 8BAlibaba Qwen · Qwen38.19B128KGQA15.3 GB9.9 GB7.5 GB2025-04
Qwen3 14BAlibaba Qwen · Qwen314.8B128KGQA27.5 GB16.7 GB11.6 GB2025-04
Qwen3 32BAlibaba Qwen · Qwen332.8B128KGQA61.0 GB33.4 GB20.5 GB2025-04
Qwen3 30B-A3BAlibaba Qwen · Qwen330.5B · 3.35B act.128KGQA56.9 GB29.6 GB16.8 GB2025-04
Qwen3 235B-A22BAlibaba Qwen · Qwen3235B · 22.2B act.256KGQA438 GB221 GB120 GB2025-04
Phi-4-reasoning 14BMicrosoft · Phi-414.7B32KGQA27.3 GB15.6 GB10.1 GB2025-04
Llama-3.1-Nemotron-Ultra-253BNVIDIA · Nemotron253B128KGQA472 GB244 GB137 GB2025-04
Granite 3.3 8BIBM · Granite 3.38.17B128KGQA15.2 GB7.8 GB4.3 GB2025-04
Gemma 3 27BGoogle · Gemma 327.4B128KGQA51.1 GB26.9 GB15.5 GB2025-03
Gemma 3 12BGoogle · Gemma 312.2B128KGQA22.7 GB12.3 GB7.4 GB2025-03
Gemma 3 4BGoogle · Gemma 34.30B128KGQA8.0 GB4.6 GB3.0 GB2025-03
Gemma 3 1BGoogle · Gemma 31.00B32KMQA1.9 GB1.2 GB0.9 GB2025-03
Command A (03-2025)Cohere · Command111B256KGQA207 GB106 GB59.3 GB2025-03
OLMo 2 32BAllenAI · OLMo 232.2B4KGQA60.0 GB31.9 GB18.8 GB2025-03
Phi-4-mini 3.8BMicrosoft · Phi-43.84B128KGQA7.1 GB4.1 GB2.7 GB2025-02
DeepSeek R1DeepSeek · DeepSeek R1684B · 37.6B act.160KMLA1,275 GB641 GB344 GB2025-01
R1-Distill-Qwen 1.5BDeepSeek · DeepSeek R1 Distill1.78B128KGQA3.3 GB2.5 GB2.2 GB2025-01
R1-Distill-Qwen 7BDeepSeek · DeepSeek R1 Distill7.62B128KGQA14.2 GB9.1 GB6.8 GB2025-01
R1-Distill-Qwen 14BDeepSeek · DeepSeek R1 Distill14.8B128KGQA27.5 GB16.7 GB11.6 GB2025-01
R1-Distill-Qwen 32BDeepSeek · DeepSeek R1 Distill32.8B128KGQA61.0 GB33.4 GB20.5 GB2025-01
R1-Distill-Llama 8BDeepSeek · DeepSeek R1 Distill8.03B128KGQA15.0 GB9.4 GB6.8 GB2025-01
R1-Distill-Llama 70BDeepSeek · DeepSeek R1 Distill70.6B128KGQA131 GB69.6 GB40.7 GB2025-01
InternLM3 8BShanghai AI Lab · InternLM8.80B32KGQA16.4 GB10.2 GB7.2 GB2025-01
Llama 3.3 70BMeta · Llama 3.370.6B128KGQA131 GB69.6 GB40.7 GB2024-12
DeepSeek V3DeepSeek · DeepSeek V3684B · 37.6B act.160KMLA1,275 GB641 GB344 GB2024-12
Phi-4 14BMicrosoft · Phi-414.7B16KGQA27.3 GB15.6 GB10.1 GB2024-12
Falcon 3 7BTII · Falcon 37.46B32KGQA13.9 GB8.4 GB5.9 GB2024-12
Falcon 3 10BTII · Falcon 310.3B32KGQA19.2 GB11.1 GB7.3 GB2024-12
Qwen2.5-Coder 32BAlibaba Qwen · Qwen2.532.8B128KGQA61.0 GB33.4 GB20.5 GB2024-11
Mistral Large 2 (2411)Mistral AI · Mistral Large123B128KGQA228 GB116 GB62.9 GB2024-11
OLMo 2 7BAllenAI · OLMo 27.30B4KMHA13.6 GB8.3 GB5.9 GB2024-11
OLMo 2 13BAllenAI · OLMo 213.7B4KMHA25.5 GB14.7 GB9.6 GB2024-11
Llama 3.2 1BMeta · Llama 3.21.24B128KGQA2.3 GB1.4 GB1.0 GB2024-09
Llama 3.2 3BMeta · Llama 3.23.21B128KGQA6.0 GB3.4 GB2.1 GB2024-09
Qwen2.5 7BAlibaba Qwen · Qwen2.57.62B128KGQA14.2 GB9.1 GB6.8 GB2024-09
Qwen2.5 14BAlibaba Qwen · Qwen2.514.8B128KGQA27.5 GB16.7 GB11.6 GB2024-09
Qwen2.5 32BAlibaba Qwen · Qwen2.532.8B128KGQA61.0 GB33.4 GB20.5 GB2024-09
Qwen2.5 72BAlibaba Qwen · Qwen2.572.7B128KGQA135 GB72.4 GB42.8 GB2024-09
Command R (08-2024)Cohere · Command32.3B128KGQA60.2 GB32.0 GB18.8 GB2024-08
Command R+ (08-2024)Cohere · Command104B128KGQA193 GB99.6 GB55.7 GB2024-08
Gemma 2 2BGoogle · Gemma 22.61B8KGQA4.9 GB3.0 GB2.1 GB2024-07
Llama 3.1 8BMeta · Llama 3.18.03B128KGQA15.0 GB9.4 GB6.8 GB2024-07
Llama 3.1 70BMeta · Llama 3.170.6B128KGQA131 GB69.6 GB40.7 GB2024-07
Llama 3.1 405BMeta · Llama 3.1406B128KGQA756 GB386 GB212 GB2024-07
Mistral Nemo 12BMistral AI · Mistral12.2B128KGQA22.8 GB13.9 GB9.7 GB2024-07
Gemma 2 27BGoogle · Gemma 227.2B8KGQA50.7 GB26.5 GB15.1 GB2024-06
Gemma 2 9BGoogle · Gemma 29.24B8KGQA17.2 GB9.5 GB5.8 GB2024-06
Mistral 7B v0.3Mistral AI · Mistral7.25B32KGQA13.5 GB7.2 GB4.3 GB2024-05
Codestral 22BMistral AI · Mistral22.2B32KGQA41.4 GB21.5 GB12.1 GB2024-05
Yi-1.5 34B01.AI · Yi34.4B4KGQA64.1 GB33.7 GB19.5 GB2024-05
Mixtral 8x22BMistral AI · Mixtral141B · 39.2B act.64KGQA262 GB132 GB70.7 GB2024-04
Mixtral 8x7BMistral AI · Mixtral46.7B · 12.9B act.32KGQA87.0 GB44.0 GB23.8 GB2023-12
Olmo Hybrid 7BAllenAI · Olmo 37.43B64Khybrid13.8 GB8.4 GB5.8 GB

Weights only, with embeddings kept in 16-bit for quantized formats. KV cache and runtime come on top: open a model for the full picture.