Coarena - Task Completion: leaderboard
Coarena agent-reported completion: share of runs with a trajectory whose terminal done call reports success, excluding harness errors. This is a self-declared outcome, separate from the blind human preference rating and the independently verified-completion metric.
Metric: Completion Rate (%). Source: coarena.ai. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Wonder | 97.8 | #16 |
| 2 | Muse Spark 1.3 | 97.7 | #31 |
| 3 | hy4 | 97.7 | #20 |
| 4 | DeepSeek V4.1 Flash (1017) | 96.9 | #28 |
| 5 | Claude Haiku 5.5 | 95.6 | #46 |
| 6 | Gemini 4 Argon | 95.5 | #9 |
| 7 | MiniMax-M3.1-Flash | 93.7 | #57 |
| 8 | GPT-6 Luna | 93.6 | #62 |
| 9 | GPT-6.1 Terra | 93.1 | #14 |
| 10 | Kimi K3.1 Pro | 92 | #18 |
| 11 | Claude Sonnet 5.5 | 91.5 | #13 |
| 12 | Claude Opus 5.5 | 85.9 | #3 |
| 13 | GPT-6.1 Sol | 85 | #6 |
| 14 | MiMo-V2.6-Pro | 81 | #51 |
| 15 | Claude Fable 5.1 | 80.7 | #10 |
Interactive version: theaggregate.ai/benchmark?slug=coarena-task-completion · How It Works · Data refreshed daily, snapshot 2026-11-07.