LoopsBench - Claude Code Loop - Resolve Rate: leaderboard
LoopsBench (arXiv 2608.00267v2, Table 2 model sweep): eight models from different vendors on 112 dependency-structured long-horizon coding tasks, all run in the Claude Code loop. Share of tasks where every released test, regression obligations included, passes at task end after outer-loop continuation rounds on the preserved workspace; percent, higher is better.
Metric: Resolve rate (%; with outer-loop continuation). Source: arxiv.org. Saturation forecast: Around January 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 25 | #51 |
| 2 | GPT-5.5 | 20.54 | #33 |
| 3 | DeepSeek V4 Pro | 18.75 | #112 |
| 4 | GLM-5.1 | 18.75 | #109 |
| 5 | Gemini 3.1 Pro (Preview) | 14.29 | #64 |
| 6 | Qwen 3.6 Plus | 9.82 | #124 |
| 7 | Kimi K2.6 | 6.25 | #113 |
| 8 | Grok 4.1 Fast (Reasoning) | 4.46 | #227 (Grok 4.1 Fast) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=loopsbench-claude-code-loop-resolve-rate · How It Works · Data refreshed daily, snapshot 2026-11-05.