Chess Puzzles (Epoch AI) (Chess Puzzles)
We show this table for reference; we do not rank on it.
Epoch AI's Chess Puzzles benchmark gives models 100 novel chess positions as FEN text and asks for the single best next move, as judged by Stockfish.
Benchmark score on Chess Puzzles — November 13, 2026
We compile the Chess Puzzles rows from benchmark-owner or independent runs. Claude Fable 5.2 leads the table at 79.0%, followed by Gemini 4 Carbon (78.0%) and GPT-6 Astra (74.0%). We do not use these results to rank models overall.
Claude Fable 5.2
Anthropic
Run by Chess PuzzlesGemini 4 Carbon
GPT-6 Astra
OpenAI
Run by Chess Puzzles87 modelsReasoningCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (87 models)
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.2AnthropicRun by Chess Puzzles | 79.0% | Run by Chess Puzzles | Not reported | Closed |
| 2 | Gemini 4 CarbonGoogleRun by Chess Puzzles | 78.0% | Run by Chess Puzzles | Not reported | Closed |
| 3 | GPT-6 AstraOpenAIRun by Chess Puzzles | 74.0% | Run by Chess Puzzles | Not reported | Closed |
| 4 | GPT-6.1 SolOpenAIRun by Chess Puzzles | 68.0% | Run by Chess Puzzles | Not reported | Closed |
| 5 | Claude Opus 5.5AnthropicRun by Chess Puzzles | 67.0% | Run by Chess Puzzles | Not reported | Closed |
| 6 | Gemini 3.7 FlashGoogleRun by Chess Puzzles | 66.0% | Run by Chess Puzzles | Not reported | Closed |
| 7 | GPT-5.5OpenAIRun by Chess Puzzles | 63.0% | Run by Chess Puzzles | Not reported | Closed |
| 8 | GPT-5.5 ProOpenAIRun by Chess Puzzles | 62.0% | Run by Chess Puzzles | Not reported | Closed |
| 9 | GPT-5.2 ProOpenAIRun by Chess Puzzles | 60.0% | Run by Chess Puzzles | Not reported | Closed |
| 10 | GPT-5.6 SolOpenAIRun by Chess Puzzles | 59.0% | Run by Chess Puzzles | Not reported | Closed |
| 11 | Claude Sonnet 5.5AnthropicRun by Chess Puzzles | 59.0% | Run by Chess Puzzles | Not reported | Closed |
| 12 | Gemini 4 ArgonGoogleRun by Chess Puzzles | 59.0% | Run by Chess Puzzles | Not reported | Closed |
| 13 | GPT-5.4 ProOpenAIRun by Chess Puzzles | 59.0% | Run by Chess Puzzles | Not reported | Closed |
| 14 | Gemini 3.8 FlashGoogleRun by Chess Puzzles | 58.0% | Run by Chess Puzzles | Not reported | Closed |
| 15 | GPT-6 SolOpenAIRun by Chess Puzzles | 56.0% | Run by Chess Puzzles | Not reported | Closed |
| 16 | Claude Haiku 5.5AnthropicRun by Chess Puzzles | 55.0% | Run by Chess Puzzles | Not reported | Closed |
| 17 | Claude Opus 4.8AnthropicRun by Chess Puzzles | 52.0% | Run by Chess Puzzles | Not reported | Closed |
| 18 | Gemini 3.1 ProGoogleRun by Chess Puzzles | 51.0% | Run by Chess Puzzles | Not reported | Closed |
| 19 | Claude Fable 5.1AnthropicRun by Chess Puzzles | 48.0% | Run by Chess Puzzles | Not reported | Closed |
| 20 | GPT-5.4OpenAIRun by Chess Puzzles | 48.0% | Run by Chess Puzzles | Not reported | Closed |
| 21 | GPT-5.2OpenAIRun by Chess Puzzles | 48.0% | Run by Chess Puzzles | Not reported | Closed |
| 22 | Claude Fable 5AnthropicRun by Chess Puzzles | 47.0% | Run by Chess Puzzles | Not reported | Closed |
| 23 | GPT-5.6 TerraOpenAIRun by Chess Puzzles | 47.0% | Run by Chess Puzzles | Not reported | Closed |
| 24 | Qwen3.8 MaxAlibabaRun by Chess Puzzles | 46.0% | Run by Chess Puzzles | Not reported | Open |
| 25 | DeepSeek V4 Pro 0813DeepSeekRun by Chess Puzzles | 46.0% | Run by Chess Puzzles | Not reported | Open |
| 26 | DeepSeek V4.1 Flash 1017DeepSeekRun by Chess Puzzles | 46.0% | Run by Chess Puzzles | Not reported | Open |
| 27 | Claude Sonnet 5AnthropicRun by Chess Puzzles | 45.0% | Run by Chess Puzzles | Not reported | Closed |
| 28 | Gemini 3.5 FlashGoogleRun by Chess Puzzles | 45.0% | Run by Chess Puzzles | Not reported | Closed |
| 29 | GPT-5.4 nanoOpenAIRun by Chess Puzzles | 45.0% | Run by Chess Puzzles | Not reported | Closed |
| 30 | Kimi K3.1 ProMoonshot AIRun by Chess Puzzles | 45.0% | Run by Chess Puzzles | Not reported | Open |
| 31 | Claude Opus 5AnthropicRun by Chess Puzzles | 43.0% | Run by Chess Puzzles | Not reported | Closed |
| 32 | GPT-5.6 LunaOpenAIRun by Chess Puzzles | 43.0% | Run by Chess Puzzles | Not reported | Closed |
| 33 | GPT-5.1OpenAIRun by Chess Puzzles | 42.0% | Run by Chess Puzzles | Not reported | Closed |
| 34 | Grok 4.7xAIRun by Chess Puzzles | 41.0% | Run by Chess Puzzles | Not reported | Closed |
| 35 | Claude Opus 4.7AnthropicRun by Chess Puzzles | 41.0% | Run by Chess Puzzles | Not reported | Closed |
| 36 | Gemini 3.6 FlashGoogleRun by Chess Puzzles | 40.0% | Run by Chess Puzzles | Not reported | Closed |
| 37 | Gemini 3 FlashGoogleRun by Chess Puzzles | 40.0% | Run by Chess Puzzles | Not reported | Closed |
| 38 | Muse Spark 1.3MetaRun by Chess Puzzles | 37.0% | Run by Chess Puzzles | Not reported | Closed |
| 39 | GLM-5.3Z.AIRun by Chess Puzzles | 37.0% | Run by Chess Puzzles | Not reported | Open |
| 40 | Kimi K3Moonshot AIRun by Chess Puzzles | 36.0% | Run by Chess Puzzles | Not reported | Open |
| 41 | o3OpenAIRun by Chess Puzzles | 36.0% | Run by Chess Puzzles | Not reported | Closed |
| 42 | GPT-5 (high)OpenAIRun by Chess Puzzles | 36.0% | Run by Chess Puzzles | Not reported | Closed |
| 43 | GPT-6 LunaOpenAIRun by Chess Puzzles | 36.0% | Run by Chess Puzzles | Not reported | Closed |
| 44 | Grok 4.5xAIRun by Chess Puzzles | 35.0% | Run by Chess Puzzles | Not reported | Closed |
| 45 | Muse Spark 1.1MetaRun by Chess Puzzles | 34.0% | Run by Chess Puzzles | Not reported | Closed |
| 46 | Grok 4.6xAIRun by Chess Puzzles | 33.0% | Run by Chess Puzzles | Not reported | Closed |
| 47 | Mistral Large 4MistralRun by Chess Puzzles | 33.0% | Run by Chess Puzzles | Not reported | Open |
| 48 | GPT-5 (medium)OpenAIRun by Chess Puzzles | 32.0% | Run by Chess Puzzles | Not reported | Closed |
| 49 | DeepSeek V4 Flash 0731DeepSeekRun by Chess Puzzles | 32.0% | Run by Chess Puzzles | Not reported | Open |
| 50 | Grok 4.20xAIRun by Chess Puzzles | 29.0% | Run by Chess Puzzles | Not reported | Closed |
| 51 | Qwen3.7 MaxAlibabaRun by Chess Puzzles | 27.0% | Run by Chess Puzzles | Not reported | Closed |
| 52 | Muse Spark 1.2MetaRun by Chess Puzzles | 27.0% | Run by Chess Puzzles | Not reported | Closed |
| 53 | Seed 2.1 TurboByteDanceRun by Chess Puzzles | 26.0% | Run by Chess Puzzles | Not reported | Closed |
| 54 | GPT-5 miniOpenAIRun by Chess Puzzles | 24.0% | Run by Chess Puzzles | Not reported | Closed |
| 55 | Claude Sonnet 4.6AnthropicRun by Chess Puzzles | 23.0% | Run by Chess Puzzles | Not reported | Closed |
| 56 | GLM-5.2Z.AIRun by Chess Puzzles | 22.0% | Run by Chess Puzzles | Not reported | Open |
| 57 | GPT-5 nanoOpenAIRun by Chess Puzzles | 22.0% | Run by Chess Puzzles | Not reported | Closed |
| 58 | InklingThinking Machines LabRun by Chess Puzzles | 21.0% | Run by Chess Puzzles | Not reported | Open |
| 59 | Claude Opus 4.6AnthropicRun by Chess Puzzles | 21.0% | Run by Chess Puzzles | Not reported | Closed |
| 60 | GLM-4.7Z.AIRun by Chess Puzzles | 21.0% | Run by Chess Puzzles | Not reported | Open |
| 61 | GLM-5.3-FlashZ.AIRun by Chess Puzzles | 21.0% | Run by Chess Puzzles | Not reported | Open |
| 62 | Seed 2.1 ProByteDanceRun by Chess Puzzles | 21.0% | Run by Chess Puzzles | Not reported | Closed |
| 63 | Kimi K2.6Moonshot AIRun by Chess Puzzles | 20.0% | Run by Chess Puzzles | Not reported | Open |
| 64 | GLM-5.1Z.AIRun by Chess Puzzles | 20.0% | Run by Chess Puzzles | Not reported | Open |
| 65 | DeepSeek V4 ProDeepSeekRun by Chess Puzzles | 19.0% | Run by Chess Puzzles | Not reported | Open |
| 66 | Grok 4.3xAIRun by Chess Puzzles | 18.0% | Run by Chess Puzzles | Not reported | Closed |
| 67 | MiniMax M3MiniMaxRun by Chess Puzzles | 17.0% | Run by Chess Puzzles | Not reported | Open |
| 68 | GPT-5.4 miniOpenAIRun by Chess Puzzles | 17.0% | Run by Chess Puzzles | Not reported | Closed |
| 69 | Qwen3.6 PlusAlibabaRun by Chess Puzzles | 16.0% | Run by Chess Puzzles | Not reported | Closed |
| 70 | Gemini 3.5 Flash-LiteGoogleRun by Chess Puzzles | 16.0% | Run by Chess Puzzles | Not reported | Closed |
| 71 | DeepSeek V4 Pro (High)DeepSeekRun by Chess Puzzles | 16.0% | Run by Chess Puzzles | Not reported | Open |
| 72 | o3-miniOpenAIRun by Chess Puzzles | 15.0% | Run by Chess Puzzles | Not reported | Closed |
| 73 | Kimi K2.7 CodeMoonshot AIRun by Chess Puzzles | 14.0% | Run by Chess Puzzles | Not reported | Open |
| 74 | Claude Opus 4.5AnthropicRun by Chess Puzzles | 14.0% | Run by Chess Puzzles | Not reported | Closed |
| 75 | Qwen3.5 FlashAlibabaRun by Chess Puzzles | 14.0% | Run by Chess Puzzles | Not reported | Closed |
| 76 | o4-mini (high)OpenAIRun by Chess Puzzles | 13.0% | Run by Chess Puzzles | Not reported | Closed |
| 77 | GPT-4.1OpenAIRun by Chess Puzzles | 11.0% | Run by Chess Puzzles | Not reported | Closed |
| 78 | o1OpenAIRun by Chess Puzzles | 10.0% | Run by Chess Puzzles | Not reported | Closed |
| 79 | Claude Sonnet 4.5AnthropicRun by Chess Puzzles | 10.0% | Run by Chess Puzzles | Not reported | Closed |
| 80 | GPT-4oOpenAIRun by Chess Puzzles | 9.0% | Run by Chess Puzzles | Not reported | Closed |
| 81 | Qwen3.5 PlusAlibabaRun by Chess Puzzles | 9.0% | Run by Chess Puzzles | Not reported | Closed |
| 82 | Claude Haiku 4.5AnthropicRun by Chess Puzzles | 8.0% | Run by Chess Puzzles | Not reported | Closed |
| 83 | Seed 1.6ByteDanceRun by Chess Puzzles | 5.0% | Run by Chess Puzzles | Not reported | Closed |
| 84 | Claude 3 OpusAnthropicRun by Chess Puzzles | 4.0% | Run by Chess Puzzles | Not reported | Closed |
| 85 | GPT-4o miniOpenAIRun by Chess Puzzles | 3.0% | Run by Chess Puzzles | Not reported | Closed |
| 86 | GPT-4.1 miniOpenAIRun by Chess Puzzles | 1.0% | Run by Chess Puzzles | Not reported | Closed |
| 87 | GPT-4.1 nanoOpenAIRun by Chess Puzzles | 0.0% | Run by Chess Puzzles | Not reported | Closed |
- Lab-reported
- Benchmark-run
- Independent
How to read this leaderboard
Higher is better. Each puzzle is worth one point, so on 100 puzzles one answer moves the score by a full point; Epoch publishes a standard error of about 4 to 5 points for frontier models. Claude Fable 5.2 leads at 79%.
Operator receipt: 87 sourced rows are currently displayable on this page; the leading published row is Claude Fable 5.2 at 79.0%.
Honest limit: Chess is a narrow skill, and Epoch presents it as a cheap proxy for spatial reasoning and planning rather than a broad capability test. Results depend on the long token budget and Epoch's agent scaffold. OpenAI Pro-mode rows are left out because BenchLM has no separate row for them. BenchLM shows Chess Puzzles as display-only; it does not enter overall or category rankings.
Among the reported Chess Puzzles rows, Claude Fable 5.2 is first at 79.0%. The third row is 5.0 points behind. The broader top-10 range is 20.0 points, so the table still separates the published systems.
87 models have been evaluated on Chess Puzzles. The benchmark falls in the Reasoning category. We keep these results separate from the weighted global scoring system, so they remain source-specific evidence. Chess Puzzles is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Chess Puzzles
Year
2026
Tasks
100 programmatically generated chess positions with a single best move
Format
Exact-match move answer from FEN text input
Difficulty
Novel chess positions, single best move
Epoch generated every position programmatically, so none appear in any other source. Games open with six random moves, continue with Stockfish at a random Elo rating, and full-strength Stockfish then picks positions with exactly one best move. Positions where the best move is an obvious capture are filtered out. Models get the board as a FEN string (no image), a budget of up to 1,000,000 output tokens, and must submit a move such as b1c3 through a tool call. The answer must match the key exactly.
Freshness and provenance
Version
Chess Puzzles 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does the Chess Puzzles benchmark test?
Whether a model can find the one best next move in a chess position it has never seen. Epoch generated the positions itself, so they cannot be memorized from puzzle sites.
Which model is best at chess puzzles?
Claude Fable 5.2 at max effort leads with 79%, followed by GPT-6 Astra at max (74%), GPT-6.1 Sol (68%) and Claude Opus 5.5 (67%).
Compare top models on Chess Puzzles
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.