Reasoning
Chess puzzles
Chess puzzles given to the model as text, scored on finding the correct moves.
Published by Epoch AI under CC BY 4.0. Compare it with other tests
Top 15
Each model at its best configuration, coloured by maker. The dashed line marks what random guessing would score.
The best model from each maker
Each maker’s highest-scoring model on this test, and its rank.
| Maker | Best model | Score | Rank |
|---|---|---|---|
 OpenAI | GPT-6 Astra (max) | 72.0% | #1 of 141 |
 Google | Gemini 3.8 Flash (high) | 61.0% | #4 of 141 |
 Anthropic | Claude Fable 5.1 (max) | 47.0% | #13 of 141 |
 DeepSeek | DeepSeek V4 Pro 0813 (max) | 47.0% | #13 of 141 |
 xAI | Grok 4.6 (high) | 40.0% | #20 of 141 |
 Alibaba | Qwen3.8 Max 0902 (xhigh) | 40.0% | #20 of 141 |
 Moonshot AI | Kimi K3 (max) | 39.0% | #24 of 141 |
 Meta AI | Muse Spark 1.3 (max) | 38.0% | #25 of 141 |
 Z.ai | GLM-5.2 (max) | 21.0% | #55 of 141 |
 MiniMax | MiniMax-M3 | 14.0% | #73 of 141 |
 ByteDance | ByteDance-Seed/Seed-OSS-36B-Instruct | 13.0% | #76 of 141 |
 NVIDIA | nemotron-3-ultra | 12.0% | #80 of 141 |
Every result
141 models, each once at its best configuration, highest score first. Models with the same score share a rank.
| Rank | Model | Maker | Score |
|---|---|---|---|
| #1 | GPT-6 Astra (max) | OpenAI | 72.0% |
| #2 | GPT-5.5 Pro Pre-release (xhigh) | OpenAI | 64.0% |
| #2 | GPT-5.6 Sol (promax) | OpenAI | 64.0% |
| #4 | Gemini 3.8 Flash (high) | Google | 61.0% |
| #4 | GPT-6.1 Sol (max) | OpenAI | 61.0% |
| #6 | GPT-5.4 Pro (xhigh) | OpenAI | 58.6% |
| #7 | Gemini 3.1 Pro Preview | Google | 55.0% |
| #8 | GPT-5.5 Pre-release (xhigh) | OpenAI | 54.0% |
| #8 | GPT-5.6 Terra (max) | OpenAI | 54.0% |
| #10 | Gemini 3.5 Flash (high) | Google | 50.0% |
| #11 | Gemini 3.1 Pro (high) | Google | 49.0% |
| #11 | GPT-5.2 (xhigh) | OpenAI | 49.0% |
| #13 | Claude Fable 5.1 (max) | Anthropic | 47.0% |
| #13 | DeepSeek V4 Pro 0813 (max) | DeepSeek | 47.0% |
| #13 | Gemini 3.7 Flash (high) | Google | 47.0% |
| #16 | GPT-5.4 (xhigh) | OpenAI | 44.0% |
| #17 | Gemini 3.6 Flash (low) | Google | 43.0% |
| #18 | Claude Opus 5 (max) | Anthropic | 42.0% |
| #19 | Claude Fable 5 (high) | Anthropic | 41.0% |
| #20 | Gemini 3 Flash Preview (high) | Google | 40.0% |
| #20 | GPT-5.6 Luna (max) | OpenAI | 40.0% |
| #20 | Grok 4.6 (high) | xAI | 40.0% |
| #20 | Qwen3.8 Max 0902 (xhigh) | Alibaba | 40.0% |
| #24 | Kimi K3 (max) | Moonshot AI | 39.0% |
| #25 | Grok 4.7 (xhigh) | xAI | 38.0% |
How it is scored
Best published score for each exact model configuration, as evaluated by Epoch AI. A suffix such as _high names the reasoning effort used. Standard errors are shown where the source reports them.
Results from Epoch AI (CC BY 4.0), as published. Model names link to their pages where the site can match them; makers come from the model’s page, or else from the publisher.