MageBench: leaderboard
Evaluates LLMs playing Magic: The Gathering via XMage engine. Elo-rated leaderboard measuring strategic card game play, blunder rate, and win rate.
Metric: Combined 1v1 Elo. Source: mage-bench.com. Source: mage-bench.com. Saturation forecast: Around March 2027. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 (Medium) | 1747 | #63 (Claude Opus 4.6) |
| 2 | GPT-5.2 (Medium) | 1737 | #110 (GPT-5.2) |
| 3 | DeepSeek V3.2 | 1682 | #205 |
| 4 | GPT-5.4 (Medium) | 1658 | #79 (GPT-5.4) |
| 5 | Gemini 3 Flash (Preview) (Medium) | 1622 | #82 (Gemini 3 Flash (Preview)) |
| 6 | O3 (Medium) | 1609 | #124 (O3) |
| 7 | Gemini 3.1 Pro (Preview) (Medium) | 1602 | #58 (Gemini 3.1 Pro (Preview)) |
| 8 | Qwen 3 Coder 480B A35B | 1601 | #279 |
| 9 | Qwen 3 235B A22B 2507 | 1594 | #260 |
| 10 | Llama 4 Maverick | 1590 | #464 |
| 11 | Gemini 3.1 Flash Lite (Preview) | 1578 | #194 |
| 12 | Gemini 2.5 Flash (Medium) | 1572 | #245 (Gemini 2.5 Flash) |
| 13 | Mistral Medium 3.1 | 1569 | #408 |
| 14 | GPT-5.2 | 1547 | #110 |
| 15 | GPT-4o Mini | 1546 | #599 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=magebench · How It Works · Data refreshed daily, snapshot 2026-10-20.