AIContenders

Will this model run on your hardware?

See which open-weight models you can run yourself — and how they do in our ranking.

We check your hardware only in your browser. Nothing is sent to our server and nothing is stored.

How we calculate

Memory = the weights file at a given quantisation (from the most popular GGUF repository on Hugging Face) + context memory + engine overhead. We pick the best quantisation that fits in GPU memory: Q8 is near-full quality, Q4 is good, below Q4 noticeably worse. When the model has to spill into RAM, we pick the lightest good one, because every gigabyte outside the GPU slows every answer. Speed is an estimate from memory bandwidth; in practice it depends on the app (Ollama, LM Studio, llama.cpp) and can differ by tens of percent.

Compare two devices

Open-weight models in the catalogue (191)

• MoonshotAI: Kimi K3 — 2780B parameters (186B active, MoE)
• Qwen: Qwen3.8 2.4T A95B — 2446B parameters (95B active, MoE)
• Meituan: LongCat 2.0 — 1776B parameters (115B active, MoE)
• DeepSeek: DeepSeek V4 Pro 0813 — 1650B parameters (107B active, MoE)
• DeepSeek: DeepSeek V4 Pro 0423 — 1599B parameters (104B active, MoE) · Q4_K_M 954.6 GB
• MoonshotAI: Kimi K2.6 — 1027B parameters (72B active, MoE)
• MoonshotAI: Kimi K2.5 — 1027B parameters (72B active, MoE) · Q4_K_M 621.2 GB
• MoonshotAI: Kimi K2.7 Code — 1027B parameters (72B active, MoE)
• MoonshotAI: Kimi K2 Thinking — 1026B parameters (72B active, MoE) · Q4_K_M 621.2 GB
• MoonshotAI: Kimi K2 0905 — 1026B parameters (72B active, MoE) · Q4_K_M 621.2 GB
• MoonshotAI: Kimi K2 0711 — 1026B parameters (72B active, MoE) · Q4_K_M 620.8 GB
• Xiaomi: MiMo-V2.6-Pro — 1024B parameters (71B active, MoE)
• Xiaomi: MiMo-V2.5-Pro — 1023B parameters (71B active, MoE) · Q4_K_M 629.6 GB
• Thinking Machines: Inkling — 952B parameters (69B active, MoE)
• Meta: Muse Spark Open — 822B parameters (42B active, MoE)
• Tencent: Hy4 preview — 780B parameters (62B active, MoE)
• DeepSeek: DeepSeek V4.1 Flash — 763B parameters (49B active, MoE)
• DeepSeek: DeepSeek V4.1 Flash 1017 — 763B parameters (49B active, MoE)
• Z.ai: GLM 5 — 754B parameters (60B active, MoE) · Q4_K_M 455.9 GB
• Z.ai: GLM 5.1 — 754B parameters (60B active, MoE) · Q4_K_M 464.5 GB
• Z.ai: GLM 5.3 — 753B parameters (60B active, MoE)
• Z.ai: GLM 5.2 — 753B parameters (60B active, MoE) · Q4_K_M 465.8 GB
• DeepSeek: DeepSeek V3.2 Exp — 685B parameters (55B active, MoE)
• DeepSeek: DeepSeek V3.2 — 685B parameters (55B active, MoE) · Q4_K_M 405.4 GB
• DeepSeek: DeepSeek V3 0324 — 685B parameters (55B active, MoE)
• DeepSeek: DeepSeek V3.1 Terminus — 685B parameters (55B active, MoE) · Q4_K_M 405.4 GB
• DeepSeek: DeepSeek V3.1 — 685B parameters (55B active, MoE) · Q4_K_M 405.4 GB
• DeepSeek: R1 0528 — 685B parameters (55B active, MoE) · Q4_K_M 409.2 GB
• DeepSeek: DeepSeek V3 — 684B parameters (55B active, MoE) · Q4_K_M 404.4 GB
• StepFun: Step 5 Preview — 604B parameters (43B active, MoE)
• NVIDIA: Nemotron 3 Ultra — 561B parameters (55B active, MoE) · Q4_K_M 359.2 GB
• Reflection: Beam 1.0 501B A23B — 501B parameters (23B active, MoE)
• Qwen: Qwen3 Coder 480B A35B — 480B parameters (35B active, MoE) · Q4_K_M 290.1 GB
• MiniMax: MiniMax-01 — 456B parameters (50B active, MoE)
• Nous: Hermes 3 405B Instruct — 406B parameters · Q4_K_M 243.1 GB
• Nous: Hermes 4 405B — 406B parameters · Q4_K_M 243.1 GB
• Qwen: Qwen3.5 397B A17B — 403B parameters (17B active, MoE) · Q4_K_M 244.1 GB
• Meta: Llama 4 Maverick — 402B parameters · Q4_K_M 242.8 GB
• Arcee AI: Trinity Large Thinking — 399B parameters (26B active, MoE) · Q4_K_M 241.8 GB
• Nex AGI: Nex-N2.5-Pro — 397B parameters (27B active, MoE) · Q4_K_M 250.8 GB
• Z.ai: GLM 4.5 — 358B parameters (35B active, MoE) · Q4_K_M 216.5 GB
• Z.ai: GLM 4.7 — 358B parameters (35B active, MoE) · Q4_K_M 216.5 GB
• Z.ai: GLM 4.6 — 357B parameters (35B active, MoE) · Q4_K_M 215.6 GB
• Z.ai: GLM 5.3 Flash — 321B parameters (25B active, MoE)
• Xiaomi: MiMo-V2.5 — 311B parameters (25B active, MoE) · Q4_K_M 190.9 GB
• Xiaomi: MiMo-V2.6-Flash — 311B parameters (25B active, MoE)
• DeepSeek: DeepSeek V4 Flash Vision Exp — 305B parameters (22B active, MoE)
• DeepSeek: DeepSeek V4 Flash 0731 — 304B parameters (22B active, MoE)
• Tencent: Hy3 preview — 299B parameters (27B active, MoE)
• Tencent: Hy3 — 299B parameters (27B active, MoE) · Q4_K_M 182.3 GB
• DeepSeek: DeepSeek V4 Flash 0423 — 291B parameters (21B active, MoE)
• Thinking Machines: Inkling Small — 266B parameters (19B active, MoE) · Q4_K_M 162.5 GB
• Qwen: Qwen3 VL 235B A22B Instruct — 236B parameters (22B active, MoE) · Q4_K_M 142.2 GB
• Qwen: Qwen3 235B A22B Instruct 2507 — 235B parameters (22B active, MoE) · Q4_K_M 142.2 GB
• Qwen: Qwen3 235B A22B Thinking 2507 — 235B parameters (22B active, MoE) · Q4_K_M 142.2 GB
• MiniMax: MiniMax M2.5 — 229B parameters (18B active, MoE) · Q4_K_M 138.3 GB
• MiniMax: MiniMax M2.1 — 229B parameters (18B active, MoE) · Q4_K_M 138.3 GB
• MiniMax: MiniMax M2 — 229B parameters (18B active, MoE) · Q4_K_M 138.3 GB
• MiniMax: MiniMax M2.7 — 229B parameters (18B active, MoE) · Q4_K_M 138.8 GB
• MiniMax: MiniMax M3.1 Flash — 220B parameters (18B active, MoE)
• StepFun: Step 3.7 Flash — 201B parameters · Q4_K_M 122.1 GB
• StepFun: Step 3.5 Flash — 199B parameters
• Qwen: Qwen3.8 Flash — 180B parameters (12B active, MoE)
• Mistral: Mixtral 8x22B Instruct — 141B parameters (40B active, MoE)
• inclusionAI: Ling 3.0 Flash — 127B parameters (8.3B active, MoE) · Q4_K_M 75.3 GB
• Qwen: Qwen3.5-122B-A10B — 125B parameters (10B active, MoE) · Q4_K_M 76.5 GB
• Mistral: Devstral 2 2512 — 125B parameters · Q4_K_M 74.9 GB
• inclusionAI: Ling 3.0 Flash VL — 125B parameters (8.1B active, MoE) · Q4_K_M 78.7 GB
• NVIDIA: Nemotron 3 Super — 124B parameters (12B active, MoE)
• Mistral: Mistral Small 4 — 119B parameters (9.5B active, MoE) · Q4_K_M 73.8 GB
• Poolside: Laguna S 2.1 — 118B parameters (10B active, MoE) · Q4_K_M 96.0 GB
• OpenAI: gpt-oss-120b — 117B parameters (9.3B active, MoE)
• Z.ai: GLM 4.5 Air — 110B parameters (12B active, MoE) · Q4_K_M 73.0 GB
• Meta: Llama 4 Scout — 109B parameters · Q4_K_M 65.4 GB
• Z.ai: GLM 4.6V — 108B parameters (12B active, MoE) · Q4_K_M 70.6 GB
• Z.ai: GLM 4.5V — 108B parameters (12B active, MoE) · Q4_K_M 70.4 GB
• Qwen: Qwen3 Next 80B A3B Instruct — 81B parameters (3.0B active, MoE) · Q4_K_M 48.5 GB
• Qwen: Qwen3 Next 80B A3B Thinking — 81B parameters (3.0B active, MoE) · Q4_K_M 48.5 GB
• Tencent: Hunyuan A13B Instruct — 80B parameters (13B active, MoE) · Q4_K_M 48.8 GB
• Qwen: Qwen3 Coder Next — 80B parameters (5.5B active, MoE) · Q4_K_M 48.5 GB
• Qwen: Qwen2.5 VL 72B Instruct — 73B parameters · Q4_K_M 47.4 GB
• Qwen2.5 72B Instruct — 73B parameters · Q4_K_M 47.4 GB
• Magnum v4 72B — 73B parameters · Q4_K_M 47.4 GB
• Sao10K: Llama 3.1 Euryale 70B v2.2 — 71B parameters · Q4_K_M 42.5 GB
• Meta: Llama 3.3 70B Instruct — 71B parameters · Q4_K_M 42.5 GB
• Nous: Hermes 3 70B Instruct — 71B parameters · Q4_K_M 42.5 GB
• Recursive: Flint — 40B parameters (2.9B active, MoE)
• TheDrummer: Skyfall 36B V2 — 37B parameters · Q4_K_M 22.4 GB
• Qwen: Qwen3.5-35B-A3B — 36B parameters (3.0B active, MoE) · Q4_K_M 22.0 GB
• Qwen: Qwen3.6 35B A3B — 36B parameters (3.0B active, MoE) · Q4_K_M 22.1 GB
• Nex AGI: Nex-N2.5-Mini — 35B parameters (2.8B active, MoE) · Q4_K_M 22.3 GB
• dolphin-2.9.1-yi-1.5-34b — 34B parameters · Q4_K_M 20.7 GB
• Laguna-XS.2 — 33B parameters (2.7B active, MoE) · Q4_K_M 64.1 GB
• Poolside: Laguna XS 2.1 — 33B parameters (2.7B active, MoE) · Q4_K_M 20.3 GB
• NVIDIA: Nemotron 3 Nano Omni — 33B parameters (3.0B active, MoE) · Q4_K_M 24.5 GB
• Qwen2.5 Coder 32B Instruct — 33B parameters · Q4_K_M 19.9 GB
• Qwen2.5-32B-Instruct — 33B parameters · Q4_K_M 19.9 GB
• DeepSeek-R1-Distill-Qwen-32B — 33B parameters · Q4_K_M 19.9 GB
• Qwen: Qwen3 32B — 33B parameters · Q4_K_M 19.8 GB
• NVIDIA: Nemotron 3.5 Lightning — 32B parameters (3.0B active, MoE)
• NVIDIA: Nemotron 3 Nano 30B A3B — 32B parameters (3.0B active, MoE) · Q4_K_M 22.4 GB
• Google: Gemma 4 31B — 31B parameters · Q4_K_M 18.3 GB
• Z.ai: GLM 4.7 Flash — 31B parameters (3.4B active, MoE) · Q4_K_M 18.3 GB
• XingChen: Xing4.0 29B A4B — 31B parameters (4.0B active, MoE) · Q4_K_M 19.0 GB
• Qwen: Qwen3 VL 30B A3B Instruct — 31B parameters (3.0B active, MoE) · Q4_K_M 18.6 GB
• Qwen: Qwen3 VL 30B A3B Thinking — 31B parameters (3.0B active, MoE) · Q4_K_M 18.6 GB
• Qwen: Qwen3 Coder 30B A3B Instruct — 31B parameters (3.0B active, MoE) · Q4_K_M 18.6 GB
• Qwen: Qwen3 30B A3B — 31B parameters (3.0B active, MoE) · Q4_K_M 18.6 GB
• Qwen: Qwen3 30B A3B Instruct 2507 — 31B parameters (3.0B active, MoE) · Q4_K_M 18.6 GB
• Cohere: North Mini Code — 30B parameters (3.3B active, MoE) · Q4_K_M 19.2 GB
• Tencent: Hy-MT2-30B-A3B — 30B parameters (3.0B active, MoE) · Q4_K_M 18.2 GB
• Meta: Muse Glimmer 30B — 30B parameters
• Qwen: Qwen3.6 27B — 28B parameters · Q4_K_M 16.8 GB
• Qwen: Qwen3.8 27B — 28B parameters · Q4_K_M 16.5 GB
• Qwen: Qwen3.5-27B — 28B parameters · Q4_K_M 16.7 GB
• Google: Gemma 3 27B — 27B parameters · Q4_K_M 16.5 GB
• Google: Gemma 2 27B — 27B parameters · Q4_K_M 16.6 GB
• PrismML: Ternary Bonsai 2 27B — 27B parameters
• Google: Gemma 4 26B A4B — 26B parameters (4.0B active, MoE) · Q4_K_M 16.9 GB
• Mistral: Voxtral Small 24B 2507 — 24B parameters · Q4_K_M 14.3 GB
• Mistral: Mistral Small 3.1 24B — 24B parameters · Q4_K_M 14.3 GB
• Mistral: Mistral Small 3.2 24B — 24B parameters · Q4_K_M 14.3 GB
• LiquidAI: LFM2.5-24B-A2B — 24B parameters (2.0B active, MoE) · Q4_K_M 14.5 GB
• Mistral: Mistral Small 3 — 24B parameters · Q4_K_M 14.3 GB
• OpenAI: gpt-oss-safeguard-20b — 22B parameters (3.6B active, MoE) · Q4_K_M 11.6 GB
• OpenAI: gpt-oss-20b — 21B parameters (3.5B active, MoE) · Q4_K_M 11.6 GB
• Reka Flash 3 — 21B parameters · Q4_K_M 13.6 GB
• DeepSeek-Coder-V2-Lite-Instruct — 16B parameters (2.2B active, MoE) · Q4_K_M 10.4 GB
• Qwen2.5-14B-Instruct — 15B parameters · Q4_K_M 9.0 GB
• Qwen2.5-Coder-14B-Instruct — 15B parameters · Q4_K_M 9.0 GB
• Qwen: Qwen3 14B — 15B parameters · Q4_K_M 9.0 GB
• Microsoft: Phi 4 — 15B parameters · Q4_K_M 9.1 GB
• Mistral: Ministral 3 14B 2512 — 14B parameters · Q4_K_M 8.2 GB
• MythoMax 13B — 13B parameters · Q4_K_M 7.9 GB
• ReMM SLERP 13B — 13B parameters · Q4_K_M 7.9 GB
• Mistral: Mistral Nemo — 12B parameters · Q4_K_M 7.5 GB
• TheDrummer: UnslopNemo 12B — 12B parameters · Q4_K_M 7.5 GB
• Google: Gemma 3 12B — 12B parameters · Q4_K_M 7.3 GB
• Meta: Llama Guard 4 12B — 12B parameters
• Ornith-1.5-9B — 9.7B parameters
• Qwen: Qwen3.5-9B — 9.7B parameters · Q4_K_M 5.7 GB
• Moondream: Moondream 3.1 — 9.3B parameters (2.0B active, MoE)
• Moondream: Moondream 3 Preview — 9.3B parameters
• Mistral: Ministral 3 8B 2512 — 8.9B parameters · Q4_K_M 5.2 GB
• IBM: Granite 4.2 8B — 8.8B parameters · Q4_K_M 5.3 GB
• Qwen: Qwen3 VL 8B Instruct — 8.8B parameters · Q4_K_M 5.0 GB
• ByteDance: UI-TARS 7B — 8.3B parameters · Q4_K_M 4.7 GB
• DeepSeek-R1-0528-Qwen3-8B — 8.2B parameters · Q4_K_M 5.0 GB
• Qwen3-8B — 8.2B parameters · Q4_K_M 5.0 GB
• Apertus-8B-Instruct-2509 — 8.1B parameters · Q4_K_M 5.1 GB
• Llama-3.1-8B-Instruct — 8.0B parameters · Q4_K_M 4.9 GB
• Sao10K: Llama 3 8B Lunaris — 8.0B parameters · Q4_K_M 4.9 GB
• Meta-Llama-3-8B-Instruct — 8.0B parameters · Q4_K_M 4.9 GB
• Hermes-3-Llama-3.1-8B — 8.0B parameters · Q4_K_M 4.9 GB
• saiga_llama3_8b — 8.0B parameters · Q4_K_M 4.9 GB
• Tencent: Hy-MT2-7B — 8.0B parameters · Q4_K_M 4.6 GB
• Qwen2.5-7B — 7.6B parameters · Q4_K_M 4.7 GB
• Qwen: Qwen2.5 7B Instruct — 7.6B parameters · Q4_K_M 4.7 GB
• Qwen2.5-Coder-7B-Instruct — 7.6B parameters · Q4_K_M 4.7 GB
• Qwen2.5-Coder-7B — 7.6B parameters · Q4_K_M 4.7 GB
• Olmo-3-7B-Think — 7.3B parameters · Q4_K_M 4.5 GB
• Mistral-7B-Instruct-v0.2 — 7.2B parameters · Q4_K_M 4.4 GB
• Reka Edge — 7.1B parameters · Q4_K_M 4.7 GB
• deepseek-coder-7b-instruct-v1.5 — 6.9B parameters · Q4_K_M 4.2 GB
• Google: Gemma 3 4B — 4.3B parameters · Q4_K_M 2.5 GB
• NVIDIA: Nemotron 3.5 Content Safety — 4.3B parameters · Q4_K_M 2.5 GB
• Qwen3-4B — 4.0B parameters · Q4_K_M 2.5 GB
• Qwen3-4B-Thinking-2507 — 4.0B parameters · Q4_K_M 2.5 GB
• Qwen3-4B-Instruct-2507 — 4.0B parameters · Q4_K_M 2.5 GB
• NVIDIA-Nemotron-3-Nano-4B-BF16 — 4.0B parameters · Q4_K_M 2.8 GB
• Mistral: Ministral 3 3B 2512 — 3.8B parameters · Q4_K_M 2.1 GB
• Phi-3.5-mini-instruct — 3.8B parameters · Q4_K_M 2.4 GB
• granite-4.1-3b — 3.4B parameters · Q4_K_M 2.1 GB
• Meta: Llama 3.2 3B Instruct — 3.2B parameters · Q4_K_M 2.0 GB
• IBM: Granite 4.0 Micro — 3.2B parameters · Q4_K_M 1.9 GB
• Qwen2.5-Coder-3B-Instruct — 3.1B parameters · Q4_K_M 2.1 GB
• Qwen2.5-3B-Instruct — 3.1B parameters · Q4_K_M 2.1 GB
• SmolLM3-3B — 3.1B parameters · Q4_K_M 1.9 GB
• LiquidAI: LFM2.5-2.6B — 2.7B parameters · Q4_K_M 1.7 GB
• MiniCPM5-2B — 2.5B parameters · Q4_K_M 1.6 GB
• Tencent: Hy-MT2-1.8B — 2.0B parameters · Q4_K_M 1.1 GB
• Qwen3-1.7B — 2.0B parameters · Q4_K_M 1.1 GB
• DeepSeek-R1-Distill-Qwen-1.5B — 1.8B parameters · Q4_K_M 1.1 GB
• Qwen2.5-1.5B-Instruct — 1.5B parameters · Q4_K_M 1.1 GB
• Qwen2.5-1.5B — 1.5B parameters · Q4_K_M 1.0 GB
• Qwen2.5-Coder-1.5B-Instruct — 1.5B parameters · Q4_K_M 1.1 GB
• Qwen2.5-Coder-1.5B — 1.5B parameters · Q4_K_M 1.0 GB
• Meta: Llama 3.2 1B Instruct — 1.2B parameters · Q4_K_M 0.8 GB
• TinyLlama-1.1B-Chat-v1.0 — 1.1B parameters · Q4_K_M 0.7 GB
• gemma-3-1b-it — 1.0B parameters · Q4_K_M 0.8 GB
• Qwen3-0.6B — 0.8B parameters