MARS-Bench - Information Reasoning: leaderboard
Each model, in an ultra multi-turn chat over the text play-by-play of 30 games (15 English NBA, 10 Chinese NBA, 5 English NHL), must track the score, count lead changes and judge a player's effect from facts spread over the dialogue; the score is the category score. A DeepSeek-V3-0324 judge awards checklist points built from the play-by-play.
Metric: Score (0-100). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked. Updated 17 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 05-06) | 70.92 | #271 |
| 2 | O1 (2024-12-17) | 62.63 | #314 |
| 3 | Doubao-1.5-Thinking-Pro | 52.17 | #257 |
| 4 | GPT-4.5 (Preview) | 50.43 | #499 |
| 5 | DeepSeek R1 | 40.01 | #487 |
| 6 | Claude 3.7 Sonnet | 36.8 | #476 |
| 7 | GPT-4o (2024-11-20) | 36.69 | #864 |
| 8 | Doubao-1.5-Pro-32k | 33.63 | #696 |
| 9 | DeepSeek V3 (1226) | 28.71 | #807 |
| 10 | Qwen 2.5 Max | 26.76 | #966 |
| 11 | GPT-4.1 Mini | 26.39 | #793 |
| 12 | GLM-Z1-Air | 24.36 | |
| 13 | Qwen 2.5 72B Instruct | 24.06 | #1119 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=mars-bench-information-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-17.