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

Metric: Pairwise accuracy (%) on the 73 hard 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 Plus68.77#828
2Qwen 3 235B A22B 2507 Instruct67.4#625
3Qwen Max66.71#1094
4InternLM2-7B-Reward66.23
5INF-ORM-Llama3.1-70B65.03
6DeepSeek R165#484
7Qwen 3 30B A3B 2507 Instruct64.66#1106
8Qwen 3 235B A22B 2507 (Thinking)64.66#471
9RRM-32B62.1#1127
10DeepSeek V3.2 Exp61.58#440
11Skywork-Reward-V2-Llama-3.1-8B61.44
12Skywork-Reward-V2-Qwen3-8B61.44
13Qwen 3 4B 2507 Instruct60.21#2240
14RM-R1-DS-Distill-Qwen-32B59.15
15M-Prometheus-14B58.75#1263

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-hard · How It Works · Data refreshed daily, snapshot 2026-11-22.