Agent Arena - Bash Recovery: leaderboard
Agent Arena submetric measuring recovery from shell or command-line failures in agent sessions.
Metric: Bash Recovery (%). Source: arena.ai. Saturation forecast: Around February 2028. 51 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 (Max) | 15.04 | #11 (Claude Sonnet 5.5) |
| 2 | Claude Opus 5.5 (Max) | 15.01 | #3 (Claude Opus 5.5) |
| 3 | Claude Opus 5 (Max) | 13.57 | #12 (Claude Opus 5) |
| 4 | Claude Fable 5.1 (Max) | 11.6 | #9 (Claude Fable 5.1) |
| 5 | Claude Fable 5 (High) | 9.6 | #16 (Claude Fable 5) |
| 6 | GPT-5.5 (xHigh) | 9.55 | #31 (GPT-5.5) |
| 7 | DeepSeek V4.1 Flash (Max) | 7.86 | #36 (DeepSeek V4.1 Flash) |
| 8 | Claude Sonnet 5 (High) | 7.38 | #57 (Claude Sonnet 5) |
| 9 | Claude Opus 4.8 (High) | 7.22 | #37 (Claude Opus 4.8) |
| 10 | Grok 4.7 (xHigh) | 7.2 | #25 (Grok 4.7) |
| 11 | GPT-6 Sol (Max) | 7.03 | #14 (GPT-6 Sol) |
| 12 | Hy4 preview | 6.51 | #38 |
| 13 | Muse Spark 1.3 (Max) | 6.08 | #24 (Muse Spark 1.3) |
| 14 | Muse Spark 1.2 (xHigh) | 5.79 | #62 (Muse Spark 1.2) |
| 15 | Grok 4.6 (xHigh) | 5.68 | #26 (Grok 4.6) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=agent-arena-bash-recovery · How It Works · Data refreshed daily, snapshot 2026-10-20.