Open weights

Only rows with openWeights=true. Null is not a yes. Quality leads; unrated sits below, never as a win. Sizes this page does not have are not guessed.

The self-host fitter skips models with no sourced size. MoE memory uses total parameters, not active. This list is the catalogue flag, not a Hugging Face scrape.

Fit these to your hardware

ModelQualityInput $/MContextLicence
Kimi K31475.9$0.6801Mkimi-k3
GLM 5.31471.4$0.2001Mglm-5.3
GLM 5.21470.4$0.3091Mmit
DeepSeek V4 Pro 04231445.2$0.4501M—
Gemma 4 31B1443.4$0.140262Kapache-2.0
Qwen3.8 27B1440.7$0.0891Mapache-2.0
DeepSeek V4 Flash 04231432.1$0.0901M—
MiniMax M31432.1$0.2801Mminimax-community
DeepSeek V3.21424.5$0.260163Kmit
gpt-oss-120b1365.4$0.030131Kapache-2.0
Phi 41216.7$0.07016Kmit
Llama 4 Scout—$0.1001.3Mllama-4-community
Ministral 3 3B 2512—$0.100131Kapache-2.0
Ministral 3 8B 2512—$0.150262Kapache-2.0
Mistral Small 4—$0.150262Kapache-2.0
Ministral 3 14B 2512—$0.200262Kapache-2.0
Llama 4 Maverick—$0.2701Mllama-4-community
Command A+—$0.300192Kapache-2.0
Muse Glimmer 30B—$0.300131K—
DeepSeek V4 Flash Vision Exp—$0.4401Mmit
Mistral Large 3 2512—$0.500262Kapache-2.0
Nemotron 3 Ultra—$0.625262Kopenmdw-1.1
Kimi K2.7 Code—$0.671262Kmodified-mit
Qwen3.8 2.4T A95B—$2.001Mqwen3.8-max
Command A—$2.50256Kcc-by-nc-4.0

14 have no independent score. They are listed after the ranked rows.

Evidence & Ask