ProactBench - Recovery: leaderboard
Metric: Pass rate (%) on the Recovery triggers (turns 8-10: grounded forward-looking value tied to an earlier detail after the user signals completion) over the 198 curated dialogues of ProactBench (about 624 rubric-scored trigger points: 201 Emergent, 232 Critical, 191 Recovery), each trigger judged Pass, Partial or Fail by a GPT-5.4 judge against a rubric written before the model answers; dialogue histories are fixed from a curation run and the model regenerates only at the triggers, sampled once at temperature 0.7; higher is better. Source: arxiv.org. Saturation forecast: Around 2028. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 37.2 |
| 2 | Claude Opus 4.7 | 22.5 |
| 3 | DeepSeek V4 Flash | 18.3 |
| 4 | O4 Mini (2025-04-16) | 16.2 |
| 5 | MiMo-V2.5-Pro | 13.1 |
| 6 | Gemini 3.1 Pro (Preview) | 11.5 |
| 7 | Gemini 2.5 Pro (Medium) | 10.5 |
| 8 | GPT-4o (2024-11-20) | 8.9 |
| 9 | Kimi K2.6 | 7.4 |
| 10 | Qwen 2.5 7B Instruct | 6.8 |
| 11 | Gemini 2.5 Flash | 5.2 |
| 12 | Qwen 3 1.7B | 4.2 |
| 13 | Llama 4 Maverick | 2.1 |
Interactive version: theaggregate.ai/benchmark?slug=proactbench-recovery · How It Works · Data refreshed daily, snapshot 2026-10-07.