Best Open-Weight AI Models
Open-weight models ranked by SI Score, with the exact license each weight release carries shown next to it.
How this list is ranked: Every model whose weights are published, ranked by SI Score. “Open weights” means downloadable weights; the license column shows what you may actually do with them.
| # | Model | SI Score | SI Score | Confidence |
|---|---|---|---|---|
| 1 | Kimi K3 Moonshot AI provisional | 71.0 | 71.0 | 100% confidence 100 percent, Full |
| 2 | MiniMax-M3 minimax provisional | 68.0 | 68.0 | 100% confidence 100 percent, Full |
| 3 | GLM-5.3 Z.ai provisional | 67.7 | 67.7 | 93% confidence 93 percent, High |
| 4 | Inkling thinkingmachines provisional | 66.5 | 66.5 | 100% confidence 100 percent, Full |
| 5 | DeepSeek V4.1 Flash DeepSeek | 66.2 | 66.2 | 69% confidence 69 percent, Medium |
| 6 | GLM-5.3-Flash Z.ai provisional | 66.2 | 66.2 | 100% confidence 100 percent, Full |
| 7 | Kimi K2 Thinking Turbo Moonshot AI provisional | 66.0 | 66.0 | 64% confidence 64 percent, Medium |
| 8 | GLM-5.2 Z.ai provisional | 65.7 | 65.7 | 100% confidence 100 percent, Full |
| 9 | Qwen3.5 397B-A17B Alibaba / Qwen provisional | 64.0 | 64.0 | 69% confidence 69 percent, Medium |
| 10 | Kimi K2.6 Moonshot AI provisional | 64.0 | 64.0 | 85% confidence 85 percent, High |
| 11 | Gemma 4 26B A4B IT Google provisional | 63.4 | 63.4 | 64% confidence 64 percent, Medium |
| 12 | GLM-5.1 Z.ai provisional | 63.3 | 63.3 | 64% confidence 64 percent, Medium |
| 13 | DeepSeek V4 Pro DeepSeek provisional | 63.1 | 63.1 | 85% confidence 85 percent, High |
| 14 | DeepSeek V4 Pro 0813 DeepSeek provisional | 63.0 | 63.0 | 69% confidence 69 percent, Medium |
| 15 | Gemma 4 31B IT Google provisional | 63.0 | 63.0 | 64% confidence 64 percent, Medium |
Task pages rank on one transparent metric, not the blended SI Score — see methodology for how pillars and confidence are computed. Missing values mean the source has not reported them.