Which model for 8 × H100 · 640 Go?
640 GB usable for the model. 113 open-weight models fit in memory.
The margin goes to context: count ~1 GB per 8k tokens for a 7B model. A small margin means a short context.
Unlocked by this setup
These models fit on no smaller setup in the catalogue (the next one down offers 160 usable GB), even in their lightest quantization.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| GLM-5.3Zhipu AI · 753B | 1478 | Q4_K_M | 474 Go | +166 Go | tight | Hugging Face ↗ |
| GLM-5.2Zhipu AI · 754B | 1475 | Q4_K_M | 475 Go | +165 Go | tight | Hugging Face ↗ |
| GLM-5Zhipu AI · 744B | 1458 | Q4_K_M | 469 Go | +171 Go | tight | Hugging Face ↗ |
| Qwen3.5 397B-A17BAlibaba · 397B | 1442 | Q8_0 | 469 Go | +171 Go | tight | Hugging Face ↗ |
| DeepSeek V4-FlashDeepSeek · 284B | 1439 | Q4_K_MFP16 also fits · 627 Go | 180 Go | +460 Go | comfortable | ollama run deepseek-v4-flash |
| Nemotron 3 Ultra 550B-A55BNVIDIA · 550B | 1426 | Q4_K_M | 347 Go | +293 Go | comfortable | Hugging Face ↗ |
| Mistral Large 3Mistral AI · 675B | 1414 | Q4_K_M | 425 Go | +215 Go | comfortable | Hugging Face ↗ |
| GLM-4.5Z.ai · 355B | 1411 | Q8_0 | 420 Go | +220 Go | comfortable | Hugging Face ↗ |
| DeepSeek-R1DeepSeek · 671B | 1398 | Q4_K_M | 423 Go | +217 Go | comfortable | ollama run deepseek-r1:671b |
| DeepSeek V3DeepSeek · 685B | 1395 | Q4_K_M | 431 Go | +209 Go | comfortable | ollama run deepseek-v3 |
| Qwen3-Coder 480BAlibaba · 480B | 1388 | Q4_K_MQ8_0 also fits · 567 Go | 303 Go | +337 Go | comfortable | ollama run qwen3-coder:480b |
| Llama 4 MaverickMeta · 400B | 1327 | Q4_K_MQ8_0 also fits · 473 Go | 253 Go | +387 Go | comfortable | ollama run llama4:maverick |
| DeepSeek V4.1-FlashDeepSeek · 552B | — | FP8 | 563 Go | +77 Go | tight | Hugging Face ↗ |
| Kimi K2.7 CodeMoonshot AI · 1000B | — | Q4_K_M | 629 Go | +11 Go | tight | Hugging Face ↗ |
Also fit on a smaller setup
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Gemma 4 31BGoogle · 31B | 1452 | FP16 | 70 Go | +570 Go | comfortable | Hugging Face ↗ |
| Gemma 4 26B-A4BGoogle · 26B | 1438 | FP16 | 59 Go | +581 Go | comfortable | Hugging Face ↗ |
| Mistral Medium 3.5Mistral AI · 128B | 1426 | FP16 | 284 Go | +356 Go | comfortable | Hugging Face ↗ |
| Qwen3 235B-A22BAlibaba · 235B | 1375 | Q4_K_MFP16 also fits · 519 Go | 149 Go | +491 Go | comfortable | ollama run qwen3:235b |
| GLM-4.5 AirZ.ai · 106B | 1373 | FP16 | 235 Go | +405 Go | comfortable | Hugging Face ↗ |
| Gemma 3 27BGoogle · 27B | 1365 | Q4_K_MFP16 also fits · 61 Go | 19 Go | +621 Go | comfortable | ollama run gemma3:27b |
| Nemotron 3 Super 120B-A12BNVIDIA · 120B | 1360 | FP16 | 266 Go | +374 Go | comfortable | Hugging Face ↗ |
| Mistral Small 3.2Mistral AI · 24B | 1356 | Q4_K_MFP16 also fits · 55 Go | 17 Go | +623 Go | comfortable | ollama run mistral-small3.2 |
| gpt-oss-120bOpenAI · 117B | 1352 | MXFP4 | 73 Go | +567 Go | comfortable | ollama run gpt-oss:120b |
| Qwen3 32BAlibaba · 32.8B | 1347 | Q4_K_MFP16 also fits · 74 Go | 23 Go | +617 Go | comfortable | ollama run qwen3:32b |
| Gemma 3 12BGoogle · 12B | 1342 | Q4_K_MFP16 also fits · 28 Go | 9 Go | +631 Go | comfortable | ollama run gemma3:12b |
| QwQ 32BAlibaba · 32.8B | 1336 | Q4_K_MFP16 also fits · 74 Go | 23 Go | +617 Go | comfortable | ollama run qwq |
| Llama 4 ScoutMeta · 109B | 1322 | Q4_K_MFP16 also fits · 242 Go | 70 Go | +570 Go | comfortable | ollama run llama4:scout |
| Gemma 3n E4BGoogle · 4B | 1318 | Q4_K_MFP16 also fits · 11 Go | 5 Go | +635 Go | comfortable | ollama run gemma3n:e4b |
| Llama 3.3 70BMeta · 70.6B | 1318 | Q4_K_MFP16 also fits · 157 Go | 46 Go | +594 Go | comfortable | ollama run llama3.3:70b |
| gpt-oss-20bOpenAI · 21B | 1317 | MXFP4 | 17 Go | +623 Go | comfortable | ollama run gpt-oss:20b |
| Nemotron 3 Nano 30B-A3BNVIDIA · 30B | 1314 | FP16 | 68 Go | +572 Go | comfortable | Hugging Face ↗ |
| Granite 4.1 8BIBM · 8B | 1304 | FP16 | 20 Go | +620 Go | comfortable | Hugging Face ↗ |
| Gemma 3 4BGoogle · 4B | 1303 | Q4_K_MFP16 also fits · 11 Go | 5 Go | +635 Go | comfortable | ollama run gemma3:4b |
| Qwen2.5 72BAlibaba · 72.7B | 1303 | Q4_K_MFP16 also fits · 162 Go | 48 Go | +592 Go | comfortable | ollama run qwen2.5:72b |
| Llama 3.1 Nemotron 70BNVIDIA · 70.6B | 1299 | Q4_K_MFP16 also fits · 157 Go | 46 Go | +594 Go | comfortable | ollama run nemotron:70b |
| Qwen2.5-Coder 32BAlibaba · 32.8B | 1271 | Q4_K_MFP16 also fits · 74 Go | 23 Go | +617 Go | comfortable | ollama run qwen2.5-coder:32b |
| Aya Expanse 32BCohere · 32.3B | 1267 | Q4_K_MFP16 also fits · 73 Go | 22 Go | +618 Go | comfortable | ollama run aya-expanse:32b |
| Command R+Cohere · 104B | 1262 | Q4_K_MFP16 also fits · 231 Go | 67 Go | +573 Go | comfortable | ollama run command-r-plus |
| Phi-4Microsoft · 14.7B | 1256 | Q4_K_MFP16 also fits · 34 Go | 11 Go | +629 Go | comfortable | ollama run phi4 |
| Phi-4 ReasoningMicrosoft · 14.7B | 1256 | Q4_K_MFP16 also fits · 34 Go | 11 Go | +629 Go | comfortable | ollama run phi4-reasoning |
| OLMo 2 32BAllen AI · 32B | 1252 | FP16 | 72 Go | +568 Go | comfortable | Hugging Face ↗ |
| CodestralMistral AI · 22.2B | 1240 | Q4_K_MFP16 also fits · 51 Go | 16 Go | +624 Go | comfortable | ollama run codestral |
| Ministral 8BMistral AI · 8.02B | 1238 | FP16 | 20 Go | +620 Go | comfortable | Hugging Face ↗ |
| Llama 3.1 8BMeta · 8.03B | 1211 | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run llama3.1:8b |
| Mixtral 8x7BMistral AI · 46.7B | 1197 | Q4_K_MFP16 also fits · 105 Go | 31 Go | +609 Go | comfortable | ollama run mixtral:8x7b |
| Llama 3.2 3BMeta · 3.21B | 1167 | Q4_K_MFP16 also fits · 9 Go | 4 Go | +636 Go | comfortable | ollama run llama3.2:3b |
| Llama 3.2 1BMeta · 1.24B | 1111 | Q4_K_MFP16 also fits · 5 Go | 3 Go | +637 Go | comfortable | ollama run llama3.2:1b |
| Mistral 7BMistral AI · 7.25B | 1110 | Q4_K_MFP16 also fits · 18 Go | 7 Go | +633 Go | comfortable | ollama run mistral:7b |
Not yet ranked by the Arena
No Arena Elo linked: recent release, model absent from the Arena, or a match we have not made yet.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Qwen3 0.6BAlibaba · 0.6B | — | Q4_K_MFP16 also fits · 3 Go | 2 Go | +638 Go | comfortable | ollama run qwen3:0.6b |
| Gemma 3 1BGoogle · 1B | — | Q4_K_MFP16 also fits · 4 Go | 3 Go | +637 Go | comfortable | ollama run gemma3:1b |
| Phi-4 MiniMicrosoft · 3.8B | — | Q4_K_MFP16 also fits · 10 Go | 4 Go | +636 Go | comfortable | ollama run phi4-mini |
| Qwen3 4BAlibaba · 4.02B | — | Q4_K_MFP16 also fits · 11 Go | 5 Go | +635 Go | comfortable | ollama run qwen3:4b |
| Gemma 4 E2BGoogle · 2B | — | FP16 | 6 Go | +634 Go | comfortable | Hugging Face ↗ |
| CodeGemma 7BGoogle · 8.5B | — | Q4_K_MFP16 also fits · 21 Go | 7 Go | +633 Go | comfortable | ollama run codegemma |
| Command R7BCohere · 8B | — | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run command-r7b |
| DeepSeek-R1 Distill 8BDeepSeek · 8.03B | — | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run deepseek-r1:8b |
| Granite 3.3 8BIBM · 8.1B | — | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run granite3.3:8b |
| Hermes 3 8BNous Research · 8.03B | — | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run hermes3:8b |
| MiniCPM5-2BOpenBMB · 2.5B | — | FP16 | 7 Go | +633 Go | comfortable | Hugging Face ↗ |
| Qwen3 8BAlibaba · 8.2B | — | Q4_K_MFP16 also fits · 20 Go | 7 Go | +633 Go | comfortable | ollama run qwen3:8b |
| Falcon 3 10BTII · 10.3B | — | Q4_K_MFP16 also fits · 25 Go | 8 Go | +632 Go | comfortable | ollama run falcon3:10b |
| Granite 4.1 3BIBM · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| Ministral 3 3BMistral AI · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| SmolLM3 3BHugging Face · 3.1B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| Mistral Nemo 12BMistral AI · 12.2B | — | Q4_K_MFP16 also fits · 29 Go | 10 Go | +630 Go | comfortable | ollama run mistral-nemo |
| Gemma 4 E4BGoogle · 4B | — | FP16 | 11 Go | +629 Go | comfortable | Hugging Face ↗ |
| Qwen3 14BAlibaba · 14.8B | — | Q4_K_MFP16 also fits · 35 Go | 11 Go | +629 Go | comfortable | ollama run qwen3:14b |
| DeepSeek-Coder-V2 LiteDeepSeek · 15.7B | — | Q4_K_MFP16 also fits · 37 Go | 12 Go | +628 Go | comfortable | ollama run deepseek-coder-v2:16b |
| StarCoder 2BigCode · 16B | — | Q4_K_MFP16 also fits · 37 Go | 12 Go | +628 Go | comfortable | ollama run starcoder2:15b |
| Devstral SmallMistral AI · 24B | — | Q4_K_MFP16 also fits · 55 Go | 17 Go | +623 Go | comfortable | ollama run devstral |
| Lucie 7BOpenLLM France · 6.7B | — | FP16 | 17 Go | +623 Go | comfortable | Hugging Face ↗ |
| Magistral SmallMistral AI · 24B | — | Q4_K_MFP16 also fits · 55 Go | 17 Go | +623 Go | comfortable | ollama run magistral |
| OLMo 3 7BAllen AI · 7B | — | FP16 | 17 Go | +623 Go | comfortable | Hugging Face ↗ |
| Apertus 8BSwiss AI · 8B | — | FP16 | 20 Go | +620 Go | comfortable | Hugging Face ↗ |
| Ministral 3 8BMistral AI · 8B | — | FP16 | 20 Go | +620 Go | comfortable | Hugging Face ↗ |
| InternLM3 8BShanghai AI Lab · 8.8B | — | FP16 | 21 Go | +619 Go | comfortable | Hugging Face ↗ |
| EuroLLM 9BUTTER Project · 9.15B | — | FP16 | 22 Go | +618 Go | comfortable | Hugging Face ↗ |
| DeepSeek-R1 Distill 32BDeepSeek · 32.8B | — | Q4_K_MFP16 also fits · 74 Go | 23 Go | +617 Go | comfortable | ollama run deepseek-r1:32b |
| Gemma 4 12BGoogle · 12B | — | FP16 | 28 Go | +612 Go | comfortable | Hugging Face ↗ |
| Ministral 3 14BMistral AI · 14B | — | FP16 | 33 Go | +607 Go | comfortable | Hugging Face ↗ |
| Qwen3.6 27BAlibaba · 27B | — | FP16 | 61 Go | +579 Go | comfortable | Hugging Face ↗ |
| Qwen3.8 27BAlibaba · 27B | — | FP16 | 61 Go | +579 Go | comfortable | Hugging Face ↗ |
| Qwen3.6 35B-A3BAlibaba · 35B | — | FP16 | 79 Go | +561 Go | comfortable | Hugging Face ↗ |
| Mistral Small 4Mistral AI · 119B | — | FP16 | 264 Go | +376 Go | comfortable | Hugging Face ↗ |
Image, video, audio and 3D
These models do not generate text, so the text Arena does not rank them. Indicative order, not a quality ranking.
| Model | Elo | Quantization | RAM required | Margin | Memory headroom | Run |
|---|---|---|---|---|---|---|
| Kokoro-82MHexgrad · 0.08B | — | FP16 | 2 Go | +638 Go | comfortable | Hugging Face ↗ |
| ChatterboxResemble AI · 0.5B | — | FP16 | 3 Go | +637 Go | comfortable | Hugging Face ↗ |
| Parakeet TDT 0.6B v3NVIDIA · 0.6B | — | FP16 | 3 Go | +637 Go | comfortable | Hugging Face ↗ |
| TripoSRStability AI · 0.5B | — | FP16 | 3 Go | +637 Go | comfortable | Hugging Face ↗ |
| Stable Fast 3DStability AI · 1B | — | FP16 | 4 Go | +636 Go | comfortable | Hugging Face ↗ |
| TRELLISMicrosoft · 1.2B | — | FP16 | 5 Go | +635 Go | comfortable | Hugging Face ↗ |
| Whisper large-v3OpenAI · 1.55B | — | FP16 | 5 Go | +635 Go | comfortable | Hugging Face ↗ |
| Dia 1.6BNari Labs · 1.6B | — | FP16 | 6 Go | +634 Go | comfortable | Hugging Face ↗ |
| Hunyuan3D 2.1Tencent · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| Orpheus 3BCanopy Labs · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| Voxtral Mini 3BMistral AI · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| YuE2-3Bm-a-p (Multimodal Art Projection) · 3B | — | FP16 | 9 Go | +631 Go | comfortable | Hugging Face ↗ |
| FLUX.2 [klein] 4BBlack Forest Labs · 4B | — | FP16 | 11 Go | +629 Go | comfortable | Hugging Face ↗ |
| TRELLIS.2Microsoft · 4B | — | FP16 | 11 Go | +629 Go | comfortable | Hugging Face ↗ |
| CogVideoX 5BZhipu AI · 5B | — | FP16 | 13 Go | +627 Go | comfortable | Hugging Face ↗ |
| Wan 2.2 5BAlibaba · 5B | — | FP16 | 13 Go | +627 Go | comfortable | Hugging Face ↗ |
| Ming-Image 0.1 DesignAnt Group · 6B | — | FP16 | 15 Go | +625 Go | comfortable | Hugging Face ↗ |
| Z-Image TurboAlibaba · 6B | — | FP16 | 15 Go | +625 Go | comfortable | Hugging Face ↗ |
| HunyuanVideo 1.5Tencent · 8.3B | — | FP16 | 20 Go | +620 Go | comfortable | Hugging Face ↗ |
| GLM-ImageZhipu AI · 9B | — | FP16 | 22 Go | +618 Go | comfortable | Hugging Face ↗ |
| Mochi 1Genmo · 10B | — | FP16 | 24 Go | +616 Go | comfortable | Hugging Face ↗ |
| FLUX.1 [dev]Black Forest Labs · 12B | — | FP16 | 28 Go | +612 Go | comfortable | Hugging Face ↗ |
| HiDream-I1HiDream · 17B | — | FP16 | 39 Go | +601 Go | comfortable | Hugging Face ↗ |
| Qwen-Image 2512Alibaba · 20B | — | FP16 | 46 Go | +594 Go | comfortable | Hugging Face ↗ |
| LTX-2.3Lightricks · 22B | — | FP16 | 50 Go | +590 Go | comfortable | Hugging Face ↗ |
| Voxtral Small 24BMistral AI · 24B | — | FP16 | 55 Go | +585 Go | comfortable | Hugging Face ↗ |
| Wan 2.2 A14BAlibaba · 27B | — | FP16 | 61 Go | +579 Go | comfortable | Hugging Face ↗ |
| FLUX.2 [dev]Black Forest Labs · 32B | — | FP16 | 72 Go | +568 Go | comfortable | Hugging Face ↗ |
| HunyuanImage 3.0Tencent · 80B | — | FP16 | 178 Go | +462 Go | comfortable | Hugging Face ↗ |
An “ollama run” command without a suffix pulls the Q4_K_M build from the Ollama library: the row then shows that quantization and its memory, and flags whether FP16 or Q8_0 would also fit. Exception: a model published in a native format (MXFP4 for gpt-oss, FP8 for DeepSeek V4.1) exists in one build only, which is what the command loads. via Hugging Face or an explicit Ollama tag.
Share this verdict
One link, the same shortlist, the same reasons.