Plan-RewardBench - Multi-Turn Planning (Easy): 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 easy multi-turn planning pairs; higher is better.

Metric: Pairwise accuracy (%) on the 109 easy multi-turn planning 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. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Qwen 3 4B 2507 Instruct75#2240
2Skywork-Reward-V2-Llama-3.1-8B73.85
3Qwen 3 30B A3B 2507 Instruct72.02#1106
4Skywork-Reward-V2-Qwen3-8B70.64
5INF-ORM-Llama3.1-70B70.31
6Qwen 3 235B A22B 2507 (Thinking)70.28#471
7Qwen 3 235B A22B 2507 Instruct69.27#625
8DeepSeek V3.2 Exp69.27#440
9FsfairX-LLaMA3-RM-v0.168.81
10RRM-32B68.45#1127
11Qwen Plus68.35#828
12DeepSeek R167.66#484
13RM-R1-DS-Distill-Qwen-32B67.33
14Kimi K2 (Thinking)66.83
15Gemini 3 Flash (Preview)66.36#156

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-multi-turn-planning-easy · How It Works · Data refreshed daily, snapshot 2026-11-22.