TextArena Balanced Subset: leaderboard
TextArena leaderboard slice for competitive text-based games, reporting Trueskill ratings for the Balanced Subset across game environments such as chess, Liar's Dice, poker, negotiation, and social-deduction games.
Metric: Trueskill (rating). Source: github.com. Source: benchmarklist.com. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude 3.5 Sonnet | 30.61 | #350 |
| 2 | Ministral 8B | 28.81 | #1040 |
| 3 | Nova Pro (v1) | 28.46 | #497 |
| 4 | Claude 3.5 Haiku | 28.14 | #568 |
| 5 | DeepSeek R1 Distill Qwen 32B | 27.78 | #655 |
| 6 | Qwen 2.5 VL 72B Instruct | 27.67 | #377 |
| 7 | Qwen 2.5 32B Instruct | 26.43 | #498 |
| 8 | Nova Micro (v1) | 26.09 | #831 |
| 9 | GPT-4o Mini | 25.58 | #599 |
| 10 | Nova Lite (v1) | 25.53 | #830 |
| 11 | Qwen Plus | 25.03 | #385 |
| 12 | Gemma 3 12B (IT) | 22.68 | #664 |
| 13 | Llama 4 Scout | 22.68 | #656 |
| 14 | Gemma 3 4B (IT) | 21.69 | #992 |
| 15 | Llama 3.2 3B Instruct | 20.98 | #1352 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=textarena-balanced-subset · How It Works · Data refreshed daily, snapshot 2026-10-20.