Open Weights AI model rankings
The Open Weights track ranks 21 AI models by the AMR aggregate, open-weights models only. Snapshot 2026-10-14, updated weekly.
Snapshot 2026-10-14 · updated weekly
| Rank | Model | Vendor | AMR Score | GPQA | Context | Price (in / out) | Speed |
|---|---|---|---|---|---|---|---|
1NEW | DeepSeek Harness | DeepSeek | 88.8 | 86.5% | 256K | $0.35 / $1.40 | 105 |
2▲1 | MiniMax H3 | MiniMax | 88.2 | 85.8% | 1M | $0.60 / $2.20 | 121 |
3— | Kimi K3 | Moonshot AI | 87.5 | 93.5% | 512K | $0.50 / $2 | 98 |
4— | Muse Wonder Full | Meta | 86.8 | 88.1% | 512K | $0.90 / $3.20 | 88 |
5— | GLM-5.3 | Zhipu AI | 85.5 | 85.2% | 256K | $0.40 / $1.50 | 118 |
6— | Muse Wonder | Meta | 85.2 | 85.9% | 256K | $0.45 / $1.60 | 126 |
7— | DeepSeek V4.5 | DeepSeek | 84.5 | 83.7% | 256K | $0.30 / $1.20 | 112 |
8▲3 | GLM-5.3 Flash | Zhipu AI | 84.1 | 80.4% | 256K | $0.12 / $0.50 | 268 |
9▲1 | Qwen3.8 Next | Alibaba | 83.6 | 82.9% | 512K | $0.50 / $1.80 | 104 |
10— | Mistral Large 3.5 | Mistral AI | 82.3 | 80.7% | 256K | $0.80 / $2.40 | 132 |
11— | Qwen3.8 27B | Alibaba | 79.8 | 76.8% | 256K | $0.15 / $0.60 | 186 |
12— | MiniMax H3 Flash | MiniMax | 78.8 | 72.6% | 512K | $0.10 / $0.40 | 358 |
13— | MiMo-V2.6-Pro | Xiaomi | 78.0 | 75.4% | 256K | $0.25 / $1 | 142 |
14— | Kimi K3 Flash | Moonshot AI | 76.0 | 70.2% | 256K | $0.14 / $0.55 | 296 |
15— | Qwen3.8 Flash | Alibaba | 74.5 | 66.8% | 256K | $0.06 / $0.25 | 412 |
16— | Gemma 4 27B | 72.0 | 67.4% | 256K | $0.18 / $0.70 | 174 | |
17NEW | Strata 2 | Strata Labs | 70.5 | 62.8% | 128K | $0.12 / $0.45 | 248 |
18NEW | Jev 1 | Jev Labs | 68.9 | 59.4% | 128K | $0.08 / $0.30 | 322 |
19NEW | Clef 2 | Clef AI | 67.2 | 56.9% | 128K | $0.10 / $0.35 | 276 |
20— | Mercury 2.5 | Inception Labs | 62.0 | 48.2% | 128K | $0.10 / $0.40 | 583 |
21NEW | Laya 1 | Laya Research | 55.4 | 38.4% | 64K | $0.05 / $0.20 | 368 |
Open Weights FAQ
Which AI model is best for Open Weights?
DeepSeek Harness leads the Open Weights track as of October 2026. The full table above ranks all 21 tracked models on this dimension, with the underlying values shown for every row.
How is the Open Weights ranking calculated?
This track ranks models by the AMR aggregate, open-weights models only. Scores are re-verified weekly against public benchmark sources and API rate cards, and the snapshot date is shown on every table so you can see exactly how fresh each value is.