Plan-RewardBench - Error Recovery: leaderboard
Metric: Pairwise accuracy (%) on the 361 robust error-recovery pairs of the Plan-RewardBench chosen versus rejected agent trajectories (shared tool registry, multi-turn context, executed tool feedback), preference labels from a Gemini 3 Pro and GPT-5.1 judge panel with human audit; discriminative reward models score each trajectory, generative reward models and LLM judges choose between the pair; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Skywork-Reward-V2-Qwen3-8B | 75.35 |
| 2 | Qwen 3 30B A3B 2507 Instruct | 74.93 |
| 3 | Qwen Plus | 73.75 |
| 4 | INF-ORM-Llama3.1-70B | 69.78 |
| 5 | GPT-5 | 69.39 |
| 6 | Qwen 3 235B A22B 2507 Instruct | 68.84 |
| 7 | Qwen 3 4B 2507 Instruct | 68.7 |
| 8 | Qwen Max | 68.63 |
| 9 | DeepSeek R1 | 68.01 |
| 10 | Gemini 3 Flash (Preview) | 67.31 |
| 11 | InternLM2-7B-Reward | 67.05 |
| 12 | DeepSeek V3.2 Exp | 66.76 |
| 13 | Qwen 3 235B A22B 2507 (Thinking) | 66.48 |
| 14 | FsfairX-LLaMA3-RM-v0.1 | 65.1 |
| 15 | Skywork-Reward-V2-Llama-3.1-8B | 65.1 |
Interactive version: theaggregate.ai/benchmark?slug=plan-rewardbench-error-recovery · How It Works · Data refreshed daily, snapshot 2026-10-07.