SWE-Misbehavior - System Interaction smolagents - Average: leaderboard
SWE-Misbehavior:fixed smolagents coding agent on System Interaction tasks; rate of boundary violations, lower is better. Rule-based code/file metrics and LLM-judged instruction/environment/resource metrics; the paper Qwen-Coder-30B label lacks an exact generation binding and remains source-qualified. Misbehavior Rate (%).
Metric: Misbehavior Rate (%). Source: arxiv.org. Saturation forecast: Around 2029. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 14.5 | #242 |
| 2 | GPT-5 | 10.59 | #104 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swe-misbehavior-system-interaction-smolagents-average · How It Works · Data refreshed daily, snapshot 2026-10-23.