RIPPLE Base Actor - SWE-bench Pro: leaderboard
RIPPLE Base Actor - SWE-bench Pro. The RIPPLE paper reports the unchanged Qwen3.5-27B actor in a SWE-agent-style editing and execution harness, with a limit of 150 tool turns and one trajectory per task. Fixed methods average three matched decoding runs. SWE-bench Verified, Pro, and Multilingual use Resolved@1; SWE-Compass uses its official task-aligned macro average. Base actor refers to the unchanged starting actor, not a newly inferred base-model checkpoint. Separately trained Recovery-SFT and RL variants, and online-critic assisted runs, are not assigned to this stock actor. Reported metric: Resolved@1 (%).
Metric: Resolved@1 (%). Source: arxiv.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3.5-27B | 50.8 |
Interactive version: theaggregate.ai/benchmark?slug=ripple-base-actor-swe-bench-pro · How It Works · Data refreshed daily, snapshot 2026-10-25.