MageBench Season 1: leaderboard
Magic: The Gathering matches between LLM agents on the XMage rules engine (28,000+ cards), 1v1 duels and multiplayer Commander, ranked by Elo; Season 1, run by Gregor Stocks, was won by Gemini 3 Pro.
Metric: Rating (self-reported). Source: benchmarklist.com. Status: saturation imminent. 34 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 (Medium) | 1737 |
| 2 | DeepSeek V3.2 | 1682 |
| 3 | GPT-5.4 (Medium) | 1658 |
| 4 | O3 (Medium) | 1609 |
| 5 | Gemini 3.1 Pro (Preview) (Medium) | 1602 |
| 6 | Qwen 3 235B A22B | 1594 |
| 7 | Llama 4 Maverick | 1590 |
| 8 | Gemini 3.1 Flash Lite | 1578 |
| 9 | Mistral Medium 3.1 | 1569 |
| 10 | Kimi K2 0905 | 1558 |
| 11 | GPT-5 (Medium) | 1536 |
| 12 | GPT-5 Mini | 1516 |
| 13 | GPT-OSS-120B (Medium) | 1516 |
| 14 | Mistral Large | 1501 |
| 15 | GPT-5 Nano | 1499 |
Interactive version: theaggregate.ai/benchmark?slug=magebench-season-1 · How It Works · Data refreshed daily, snapshot 2026-09-05.