Agent Arena: leaderboard
Arena.ai agent leaderboard measuring net improvement on real-world tool orchestration sessions with success, steerability, recovery, and hallucination metrics.
Metric: Net Improvement (%). Source: arena.ai. Saturation forecast: Around November 2027. 51 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (Max) | 15.53 | #3 (Claude Opus 5.5) |
| 2 | GPT-6 Astra (Max) | 13.09 | #6 (GPT-6 Astra) |
| 3 | Claude Fable 5.1 (Max) | 12.66 | #9 (Claude Fable 5.1) |
| 4 | Claude Sonnet 5.5 (Max) | 11.95 | #11 (Claude Sonnet 5.5) |
| 5 | GPT-6.1 Sol (Max) | 11.72 | #7 (GPT-6.1 Sol) |
| 6 | GPT-6 Sol (Max) | 9.92 | #14 (GPT-6 Sol) |
| 7 | Gemini 4 Argon (High) | 9.26 | #8 (Gemini 4 Argon) |
| 8 | Claude Fable 5 (High) | 8.94 | #16 (Claude Fable 5) |
| 9 | Claude Opus 5 (Max) | 8.12 | #12 (Claude Opus 5) |
| 10 | Claude Opus 4.8 (High) | 7.13 | #37 (Claude Opus 4.8) |
| 11 | GPT-5.6 Sol (xHigh) | 6.23 | #19 (GPT-5.6 Sol) |
| 12 | Claude Sonnet 5 (High) | 4.91 | #57 (Claude Sonnet 5) |
| 13 | GPT-5.5 (xHigh) | 4.38 | #31 (GPT-5.5) |
| 14 | Muse Spark 1.3 (Max) | 4.04 | #24 (Muse Spark 1.3) |
| 15 | DeepSeek V4.1 Flash (Max) | 3.8 | #36 (DeepSeek V4.1 Flash) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=agent-arena · How It Works · Data refreshed daily, snapshot 2026-10-20.