LLM Live - Chess - Move Accuracy: leaderboard
Move quality in LLM Live's fixed-cap chess arena, measured with Stockfish depth 14 and the Lichess accuracy formula on a 0–100 scale. Reasoning is capped at 1,024 tokens per move; illegal/no-move forfeits and provider-error exclusions follow the native arena policy. Exact provider routes, game counts and provisional status are retained. Accuracy and Glicko-2 are related outcomes of the same games, not independent experiments.
Metric: Lichess move accuracy (%) at Stockfish depth 14. Source: llmlive.ai. Saturation forecast: Rough model projection: around 2026. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.7 Flash (LLM Live Run: chess;reasoning-cap=1024;provider-route=google/gemini-3.7-flash) | 96.1 | |
| 2 | Grok 4.3 (LLM Live Run: chess;reasoning-cap=1024;provider-route=x-ai/grok-4.3) | 76.5 | |
| 3 | GPT-6 Luna (LLM Live Run: chess;reasoning-cap=1024;provider-route=openai/gpt-6-luna) | 71.2 | |
| 4 | Claude Haiku 5.5 (LLM Live Run: chess;reasoning-cap=1024;provider-route=anthropic/claude-haiku-5.5) | 69.6 | |
| 5 | Kimi K2.6 (LLM Live Run: chess;reasoning-cap=1024;provider-route=moonshotai/kimi-k2.6) | 62.8 | |
| 6 | DeepSeek V4.1 Flash (LLM Live Run: chess;reasoning-cap=1024;provider-route=deepseek/deepseek-v4.1-flash) | 62.2 | |
| 7 | Qwen 3.8 Flash (LLM Live Run: chess;reasoning-cap=1024;provider-route=qwen/qwen3.8-flash) | 60.9 | |
| 8 | GLM-5.3 Flash (LLM Live Run: chess;reasoning-cap=1024;provider-route=z-ai/glm-5.3-flash) | 60.7 | |
| 9 | MiniMax-M3.1-Flash (LLM Live Run: chess;reasoning-cap=1024;provider-route=minimax/minimax-m3.1-flash) | 57.7 | |
| 10 | Mistral Small 4 (LLM Live Run: chess;reasoning-cap=1024;provider-route=mistralai/mistral-small-2603) | 55.1 |
Interactive version: theaggregate.ai/benchmark?slug=llm-live-chess-move-accuracy · How It Works · Data refreshed daily, snapshot 2026-11-05.