Plan-RewardBench - Tool Irrelevance: 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 the pairs where the tools are irrelevant or unavailable; higher is better.

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 June 2027. 21 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Flash (Preview)75.55#152
2GPT-569.73#178
3RM-R1-DS-Distill-Qwen-32B69.1
4Qwen 3 235B A22B 2507 (Thinking)67.36#466
5INF-ORM-Llama3.1-70B66.91
6Skywork-Reward-V2-Qwen3-8B65.45
7Skywork-Reward-V2-Llama-3.1-8B64.91
8M-Prometheus-14B63.89#1255
9Qwen Plus63.73#820
10InternLM2-7B-Reward62.91
11RRM-32B61.15#1115
12Qwen 3 4B 2507 Instruct60.27#2221
13DeepSeek V3.2 Exp60#439
14Qwen Max60#1092
15Qwen 3 235B A22B 2507 Instruct59.82#616

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-tool-irrelevance · How It Works · Data refreshed daily, snapshot 2026-11-21.