LoopArena - Contract Selection: leaderboard
Part of LoopArena (Alibaba DreamX, 2026): 90 execution-validated questions asking which of four candidate instructions a controller should issue next to a coding agent at a frozen point of a real run. Accuracy in percent, higher is better; invalid answers count as wrong.
Metric: Contract accuracy (%; 90 execution-validated questions asking which of four candidate Loop Contracts should be issued next from a frozen run state (Type I), one response per question, invalid responses counting as wrong; no Worker execution at evaluation time; provider-default reasoning). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 87.78 | #49 |
| 2 | DeepSeek V4 Flash (0731) | 77.78 | #139 |
| 3 | Claude Opus 4.8 | 76.67 | #56 |
| 4 | GLM-5.2 | 74.44 | #138 |
| 5 | Qwen 3.7 Plus | 72.22 | #207 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=looparena-contract-selection · How It Works · Data refreshed daily, snapshot 2026-11-13.