ParaRecover - Level 2 (Recovery): leaderboard
ParaRecover Level 2: recover from the injected error and complete the parallel tool-use task; the score averages the SI, DR and ES sub-metrics. Higher is better. Source: arxiv.org/abs/2609.12345.
Metric: Average score (%, SI/DR/ES). Source: arxiv.org. Saturation forecast: Around February 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 66.85 | #67 |
| 2 | DeepSeek V4 Pro | 66.5 | #112 |
| 3 | DeepSeek V4 Flash | 66.36 | #155 |
| 4 | Qwen 3.5 122B A10B | 65.08 | #184 |
| 5 | Claude Sonnet 4.6 | 61.9 | #94 |
| 6 | GPT-5.2 | 59.2 | #115 |
| 7 | Gemini 3.1 Pro (Preview) | 57.49 | #64 |
| 8 | GLM-4.6 | 54.86 | #265 |
| 9 | GLM 4.5 Air | 53.14 | #403 |
| 10 | Qwen 3.5 9B | 52.98 | #391 |
| 11 | Qwen 3 32B | 45.64 | #449 |
| 12 | GPT-4o | 44.74 | #357 |
| 13 | Qwen 3 8B | 41.07 | #678 |
| 14 | GPT-5 Nano | 38.23 | #439 |
| 15 | GPT-4o Mini | 36.9 | #598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pararecover-level-2-recovery · How It Works · Data refreshed daily, snapshot 2026-11-05.