MARS-Bench: leaderboard
Each model follows an ultra multi-turn chat over text play-by-play of 120 games (60 English NBA, 40 Chinese NBA, 20 English NHL) and answers questions inserted between turns; the score is the mean of four category scores of 30 games each. A DeepSeek-V3-0324 judge awards checklist points built from the play-by-play. Unlike MT-Bench's short dialogues, the questions depend on facts many turns back.
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) | 72.44 | #271 |
| 2 | O1 (2024-12-17) | 59.62 | #314 |
| 3 | GPT-4.5 (Preview) | 53.33 | #499 |
| 4 | Doubao-1.5-Thinking-Pro | 52.62 | #257 |
| 5 | DeepSeek R1 | 45.42 | #487 |
| 6 | Claude 3.7 Sonnet | 41.21 | #476 |
| 7 | Doubao-1.5-Pro-32k | 38.88 | #696 |
| 8 | GPT-4o (2024-11-20) | 35.83 | #864 |
| 9 | DeepSeek V3 (1226) | 33.16 | #807 |
| 10 | GPT-4.1 Mini | 31.23 | #793 |
| 11 | Qwen 2.5 Max | 30.41 | #966 |
| 12 | Qwen 2.5 72B Instruct | 29.21 | #1119 |
| 13 | GLM-Z1-Air | 25.84 |
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 · How It Works · Data refreshed daily, snapshot 2026-11-17.