CollabBench Cook - Game Score: leaderboard
Task efficiency in the Cook-MultiPlayer game of CollabBench (2026): the final score of an Overcooked-style cooperative kitchen, when the evaluated model drives one cook through the ProAgent framework. Higher is better. The evaluated model plays one seat of a two-player game with a simulated partner whose personality varies.
Metric: Final game score of the two-agent kitchen team (completed dish orders), Cook-MultiPlayer (Overcooked-style cooperative cooking, five kitchen scenarios): the evaluated model drives Agent 1 inside the ProAgent framework while a DeepSeek-V3.1 simulator role-plays a partner with a Big Five personality profile; mean over all evaluation trajectories; higher is better. Source: arxiv.org. Saturation forecast: Around March 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3.1 | 136.53 | #497 |
| 2 | Qwen 2.5 72B Instruct | 135.47 | #1092 |
| 3 | GPT-5.2 | 135.2 | #176 |
| 4 | Qwen 2.5 7B Instruct | 86.93 | #2564 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=collabbench-cook-game-score · How It Works · Data refreshed daily, snapshot 2026-11-13.