Game Agent Coding League: leaderboard
LLMs generate code agents for eight games: Battleship, LieOnce, Wizard, Backgammon, Connect4, WordMatrix, MiniChess and Surround Morris. Official weighted normalized tournament points preserve the official best-agent selection protocol.
Metric: Weighted normalized points (0-100); eight games, official best-agent protocol. Source: gameagentcodingleague.com. Source: gameagentcodingleague.com. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 73.21 | #31 |
| 2 | GPT-5.4 | 71.93 | #79 |
| 3 | Claude Opus 4.7 | 66.96 | #49 |
| 4 | Gemini 3.1 Pro (Preview) | 62.79 | #58 |
| 5 | Claude Opus 4.6 | 62.32 | #63 |
| 6 | GLM-5.1 | 61.39 | #97 |
| 7 | Kimi K2.6 | 61.09 | #104 |
| 8 | Claude Sonnet 4.6 | 59.71 | #90 |
| 9 | MiMo-V2.5-Pro | 58.22 | #158 |
| 10 | DeepSeek V4 Pro | 57.74 | #102 |
| 11 | Qwen 3.6 Plus | 53.16 | #120 |
| 12 | Gemini 3 Flash (Preview) | 51.55 | #82 |
| 13 | Qwen 3.6 Max Preview | 49.26 | #85 |
| 14 | GPT-5.4 Mini | 49.09 | #212 |
| 15 | DeepSeek V4 Flash | 46.02 | #137 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=game-agent-coding-league · How It Works · Data refreshed daily, snapshot 2026-10-20.