Atlas

Benchmarks

← All benchmarks

ChessBench — Hanging Piece, Historical Output Repair

Games · 2026-03-25

The 30-puzzle Hanging Piece track under ChessBench's historical protocol permitting one follow-up request after a missing or unparseable output; repaired exact UCI lines count as correct. Retained separately from single-request runs.

Top models (higher is better)

ModelScore
Gemini 3.7 Flash63.3
GPT-6 Astra56.7
Grok 4.1 Fast53.3
Grok 4.553.3
Claude Fable 550.0
Gemini 3.1 Pro Preview50.0
Gemini 3.6 Flash46.7
Grok 4.2043.3
Grok Build 0.143.3
Gemini 3.5 Flash40.0
Grok 4.336.7
Gemini 3.1 Flash Image Preview33.3
GPT-5.6 Terra30.0
GPT-5.423.3
Qwen3.6 Plus Preview23.3
Claude Opus 520.0
GPT-5.520.0
GPT-5.6 Sol20.0
GLM-5.116.7
Opus 4.713.3
Opus 4.813.3
GPT-5.6 Luna10.0
Qwen3.6 Plus (2026-04-02)10.0
Opus 4.66.7
Sonnet 4.66.7
GLM-56.7
GLM-5.23.3
Sonnet 50.0
Gemini 2.5 Pro0.0
Loading Atlas data…