Plan-RewardBench: 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, the macro-average over the seven scenarios; higher is better.

Metric: Pairwise accuracy (%), macro-average of the seven scenario columns 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. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen Plus69.96#828
2DeepSeek V3.2 Exp69.57#440
3Qwen 3 235B A22B 2507 Instruct69.54#625
4INF-ORM-Llama3.1-70B69.21
5Gemini 3 Flash (Preview)69.07#156
6DeepSeek R168.95#484
7Qwen Max68.92#1094
8Qwen 3 235B A22B 2507 (Thinking)68.6#471
9GPT-568.54#185
10Qwen 3 4B 2507 Instruct67.4#2240
11InternLM2-7B-Reward66.96
12Qwen 3 30B A3B 2507 Instruct66.75#1106
13Skywork-Reward-V2-Qwen3-8B66.46
14RRM-32B66.45#1127
15Skywork-Reward-V2-Llama-3.1-8B65.88

No result here: #2 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=plan-rewardbench · How It Works · Data refreshed daily, snapshot 2026-11-22.