PC Hardware · GPU tools
Best GPU for every open model
Pick the model you want to run. Each page ranks the cards that actually fit it — from the model's real layer count, KV cache and quantisation maths, not a vibe — and shows what renting the same card costs right now. 53 models tracked.
Qwen
- Best GPU for Qwen3 4B Instruct 2507 →4B · 256k context
- Best GPU for Qwen2.5 7B Instruct →7.6B · 32k context
- Best GPU for Qwen3 8B →8.2B · 40k context
- Best GPU for Qwen3.5 9B →9.7B · 256k context
- Best GPU for Qwen3 14B →14.8B · 40k context
- Best GPU for Qwen2.5-Coder 14B →14.8B · 32k context
- Best GPU for Qwen3.6 27B →27.8B · 256k context
- Best GPU for Qwen3.8 27B →27.8B · 256k context
- Best GPU for Qwen3.5 27B →27.8B · 256k context
- Best GPU for Qwen3 30B-A3B (MoE) →30.5B · 3.3B active · MoE · 40k context
- Best GPU for Qwen3-Coder 30B-A3B (MoE) →30.5B · MoE · 256k context
- Best GPU for Qwen3 32B →32.8B · 40k context
- Best GPU for Qwen2.5 Coder 32B →32.8B · 32k context
- Best GPU for Qwen2.5 32B Instruct →32.8B · 32k context
- Best GPU for Qwen3.6 35B-A3B (MoE) →36B · 3B active · MoE · 256k context
- Best GPU for Qwen3.5 35B-A3B (MoE) →36B · 3B active · MoE · 256k context
- Best GPU for Qwen3-Coder-Next 80B-A3B (MoE) →79.7B · 3B active · MoE · 256k context
- Best GPU for Qwen3-Next 80B-A3B (MoE) →80B · MoE · 256k context
- Best GPU for Qwen3.5 122B-A10B (MoE) →125.1B · 10B active · MoE · 256k context
- Best GPU for Qwen3.8-Flash-Next (MoE) →180B · 6B active · MoE · 256k context
- Best GPU for Qwen3 235B-A22B (MoE) →235.1B · 22B active · MoE · 256k context
Gemma
- Best GPU for Gemma 4 E2B →5.1B · 128k context
- Best GPU for Gemma 4 E4B →8B · 128k context
- Best GPU for Gemma 2 9B →9.2B
- Best GPU for Gemma 4 12B →12B · 256k context
- Best GPU for Gemma 3 12B →12.2B · 128k context
- Best GPU for Gemma 4 26B-A4B (MoE) →25.8B · 3.8B active · MoE · 256k context
- Best GPU for Gemma 2 27B →27.2B
- Best GPU for Gemma 3 27B →27.4B
- Best GPU for Gemma 4 31B →31.3B · 256k context
DeepSeek
- Best GPU for DeepSeek R1 Distill 14B →14.8B · 128k context
- Best GPU for DeepSeek R1 Distill 32B →32.8B · 128k context
- Best GPU for DeepSeek R1 Distill 70B →70.6B · 128k context
- Best GPU for DeepSeek V4.1 Flash (MoE) →763.2B · 16B active · MoE · 1024k context
- Best GPU for DeepSeek V4 Pro (MoE) →1598.8B · 49B active · MoE · 1024k context
Llama
Mistral
GLM
gpt-oss
MiniMax
Nemotron
Kimi
Ornith
Phi
Missing a model? The catalogue is the same one behind cost to run; new open-weight releases are added from their published config, and every page recomputes live.