Plan-RewardBench - Tool Irrelevance: leaderboard
Metric: Pairwise accuracy (%) on the 275 tool-irrelevance or tool-unavailable 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 March 2028. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 75.55 |
| 2 | GPT-5 | 69.73 |
| 3 | Qwen 3 235B A22B 2507 (Thinking) | 67.36 |
| 4 | INF-ORM-Llama3.1-70B | 66.91 |
| 5 | Skywork-Reward-V2-Qwen3-8B | 65.45 |
| 6 | Skywork-Reward-V2-Llama-3.1-8B | 64.91 |
| 7 | Qwen Plus | 63.73 |
| 8 | InternLM2-7B-Reward | 62.91 |
| 9 | Qwen 3 4B 2507 Instruct | 60.27 |
| 10 | DeepSeek V3.2 Exp | 60 |
| 11 | Qwen Max | 60 |
| 12 | Qwen 3 235B A22B 2507 Instruct | 59.82 |
| 13 | DeepSeek R1 | 59.09 |
| 14 | Qwen 3 30B A3B 2507 Instruct | 57.82 |
| 15 | QRM-Gemma-2-27B | 56.85 |
Interactive version: theaggregate.ai/benchmark?slug=plan-rewardbench-tool-irrelevance · How It Works · Data refreshed daily, snapshot 2026-10-07.