Plan-RewardBench - Safety Refusal: leaderboard
Plan-RewardBench (2026) tests reward models and LLM judges on preferring the better of two tool-using agent trajectories, with preference labels from a judge panel audited by humans. This board is pairwise accuracy in percent on pairs that test appropriate safety refusal, including avoiding over-refusal on benign requests; higher is better.
Metric: Pairwise accuracy (%) on the 51 safety-refusal 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: Around February 2027. 21 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 84.8 | #178 |
| 2 | Qwen 3 235B A22B 2507 (Thinking) | 78.92 | #466 |
| 3 | Kimi K2 (Thinking) | 78.63 | |
| 4 | Gemini 3 Flash (Preview) | 78.43 | #152 |
| 5 | DeepSeek V3.2 Exp | 75 | #439 |
| 6 | DeepSeek R1 | 72.55 | #482 |
| 7 | Qwen Max | 71.08 | #1092 |
| 8 | Qwen 3 235B A22B 2507 Instruct | 65.69 | #616 |
| 9 | M-Prometheus-14B | 62.44 | #1255 |
| 10 | RRM-32B | 60.3 | #1115 |
| 11 | INF-ORM-Llama3.1-70B | 58.53 | |
| 12 | Qwen 3 4B 2507 Instruct | 57.35 | #2221 |
| 13 | Skywork-Reward-V2-Qwen3-8B | 56.86 | |
| 14 | RM-R1-DS-Distill-Qwen-32B | 56.62 | |
| 15 | Qwen Plus | 55.88 | #820 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=plan-rewardbench-safety-refusal · How It Works · Data refreshed daily, snapshot 2026-11-21.