Atlas

Benchmarks

← All benchmarks

ChessBench — Mate in 1

Games · 2026-03-22

ChessBench Mate in 1 contains 30 fixed Lichess positions in which the model must return the exact UCI move that checkmates immediately. The track is stratified evenly across 800–1200, 1200–1600, and 1600–2000 puzzle-rating buckets.

Top models (higher is better)

ModelScore
Grok 4.596.7
Grok Build 0.196.7
Gemini 3.5 Flash93.3
GPT-5.6 Sol93.3
Grok 4.1 Fast93.3
Claude Fable 590.0
Gemini 3.1 Flash Image Preview90.0
GPT-5.590.0
Grok 4.2090.0
Grok 4.390.0
GPT-5.6 Luna Pro86.7
GPT-5.6 Terra86.7
Qwen3.6 Plus (2026-04-02)86.7
Gemini 3.1 Pro Preview83.3
Gemini 3 Flash Preview83.3
GPT-5.476.7
Opus 4.760.0
Opus 4.860.0
Sonnet 556.7
Qwen3.6 Plus Preview56.7
Opus 4.646.7
GLM-5.246.7
Sonnet 4.640.0
GLM-5.133.3
GLM-533.3
Haiku 4.530.0
Gemini 2.5 Pro16.7
Loading Atlas data…