Chess Bench LLM: leaderboard

Models play full chess games against other LLMs and engine anchors, receiving FEN, ASCII board, move history, prior reasoning and legal-move retry feedback. Native Glicko-2 ratings are calibrated to Lichess Classical anchors, with no minimum floor and valid negative ratings; these are not observed human Lichess account ratings or FIDE estimates. Public API stable player IDs preserve run effort labels and avoid retired/medal display badges. Game count, RD, confidence intervals, engine/anchor flags and historical frozen status are retained. Raw engine anchors remain available to existing human-reference consumers, while authoritative model exclusions keep non-LLM bots out of capability fitting.

Metric: Lichess-Classical-calibrated Glicko-2 rating (native signed points). Source: chessbenchllm.onrender.com. Saturation forecast: Around December 2027. 136 models tracked.

Top models

#ModelScoreOverall rank
1Best Expert2236
2Gemini 3.1 Pro (Preview) (High)2144#58 (Gemini 3.1 Pro (Preview))
3Gemini 3.1 Pro (Preview) (Medium)2028#58 (Gemini 3.1 Pro (Preview))
4Gemini 4 Argon (High)2022#8 (Gemini 4 Argon)
5GPT-6 Astra (High)1975#6 (GPT-6 Astra)
6Gemini 3 Pro (Preview) (High)1953#66 (Gemini 3 Pro (Preview))
7Gemini 3.5 Flash (Medium)1944#60 (Gemini 3.5 Flash)
8Gemini 3.6 Flash (Medium)1916#51 (Gemini 3.6 Flash)
9Gemini 3.8 Flash (High)1877#21 (Gemini 3.8 Flash)
10Gemini 3 Flash (Preview) (Medium)1857#82 (Gemini 3 Flash (Preview))
11GPT-6 Astra (Medium)1779#6 (GPT-6 Astra)
12Gemini 3 Flash (Preview) (High)1750#82 (Gemini 3 Flash (Preview))
13Median Expert1542
14GPT-5.6 Sol (High)1532#19 (GPT-5.6 Sol)
15GPT-6.1 Sol (High)1490#7 (GPT-6.1 Sol)

No result here: #10 Muse Wonder Full.

Interactive version: theaggregate.ai/benchmark?slug=chess-bench-llm · How It Works · Data refreshed daily, snapshot 2026-10-20.