ChessBench LLM: leaderboard
Evaluates LLM chess ability using Glicko-2 ratings calibrated with engine anchors. Tracks win/loss/draw records, move legality percentage, and cost per game.
Metric: Lichess Rating. Source: chessbenchllm.onrender.com. Status: years away from saturation. 110 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) (High) | 2081 |
| 2 | Gemini 3.1 Pro (Preview) (Medium) | 2012 |
| 3 | Gemini 3.5 Flash (Medium) | 1920 |
| 4 | Gemini 3 Pro (Preview) (High) | 1914 |
| 5 | Gemini 3.6 Flash (Medium) | 1880 |
| 6 | Gemini 3 Flash (Preview) (Medium) | 1850 |
| 7 | Gemini 3 Flash (Preview) (High) | 1744 |
| 8 | GPT-5.6 Sol (High) | 1530 |
| 9 | GPT-5.6 Luna (xHigh) | 1469 |
| 10 | GPT-5.6 Sol (xHigh) | 1465 |
| 11 | GPT-5.1 (High) | 1401 |
| 12 | Grok 4.1 Fast | 1400 |
| 13 | GPT-5.6 Sol (Medium) | 1394 |
| 14 | O3 | 1333 |
| 15 | DeepSeek V4 Flash (0731) (High) | 1304 |
Interactive version: theaggregate.ai/benchmark?slug=chessbench-llm · How It Works · Data refreshed daily, snapshot 2026-09-05.