M3-Bench - L2: leaderboard
Evaluates LLM agents' social behaviour in mixed-motive games on M3-Bench (Xie et al., 2026): 24 game-theoretic tasks across four progressive levels, each played against a fixed opponent pool with and without free-form messaging. Scores are the BTA (Behavioral Trajectory Analysis) view: rule-based indicators computed from each agent's logged actions and payoffs, normalised within the task and reported on a 0-100 scale. The scale is normative, so cooperation, fairness and joint payoff efficiency raise it and exploitation lowers it. L2 is the Level 2 average, covering long-term-relationship games that are two-player with temporal dependence, computed over all six Level 2 tasks rather than only the three shown as columns. The paper's RPA (reasoning) and CCA (communication) views are LLM-judge scores of the same episodes and are not included. This series reports M3-Bench - L2 using BTA Score.
Metric: BTA Score. Source: huggingface.co. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 (Non-reasoning) | 89 | #145 (GPT-5.1) |
| 2 | GPT-5 Pro | 89 | #89 |
| 3 | Claude Opus 4.5 | 88.8 | #87 |
| 4 | GPT-OSS-120B | 85.3 | #346 |
| 5 | DeepSeek V3.2 (Non-reasoning) | 84.2 | #210 (DeepSeek V3.2) |
| 6 | Kimi K2 (Thinking) | 81.2 | #252 (Kimi K2) |
| 7 | Mistral Large 3 | 77.8 | #410 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=m3-bench-l2 · How It Works · Data refreshed daily, snapshot 2026-11-05.