Plan-RewardBench - Multi-Turn Planning (Hard): leaderboard
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.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen Plus | 68.77 |
| 2 | Qwen 3 235B A22B 2507 Instruct | 67.4 |
| 3 | Qwen Max | 66.71 |
| 4 | InternLM2-7B-Reward | 66.23 |
| 5 | INF-ORM-Llama3.1-70B | 65.03 |
| 6 | DeepSeek R1 | 65 |
| 7 | Qwen 3 30B A3B 2507 Instruct | 64.66 |
| 8 | Qwen 3 235B A22B 2507 (Thinking) | 64.66 |
| 9 | DeepSeek V3.2 Exp | 61.58 |
| 10 | Skywork-Reward-V2-Qwen3-8B | 61.44 |
| 11 | Skywork-Reward-V2-Llama-3.1-8B | 61.44 |
| 12 | Qwen 3 4B 2507 Instruct | 60.21 |
| 13 | FsfairX-LLaMA3-RM-v0.1 | 53.42 |
| 14 | QRM-Gemma-2-27B | 49.08 |
| 15 | Kimi K2 (Thinking) | 48.77 |
Interactive version: theaggregate.ai/benchmark?slug=plan-rewardbench-multi-turn-planning-hard · How It Works · Data refreshed daily, snapshot 2026-10-07.