AdaPlanBench: leaderboard

Metric: Accuracy (%): share of tasks whose final plan satisfies every world and user constraint and passes the rubric threshold on every planning dimension (early stops and turn-limit runs fail), 307 household planning tasks with progressively revealed world and user constraints at the mid-level constraint profile: a hidden constraint is disclosed only when a proposed plan violates it (GPT-5.4 constraint judges), and the agent re-plans within the turn budget; temperature 0 (GPT-5 family at its fixed default, mean of three runs); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.

Top models

#ModelScore
1GPT-567.75
2GPT-5 Mini61.89
3Gemini 3 Flash43.32
4GPT-5 Nano42.35
5DeepSeek V4 Flash35.53
6Gemini 3.1 Pro (Preview)34.53
7Llama 3.3 70B Instruct29.32
8Qwen 3 32B17.92
9Qwen 3 14B17.26
10Qwen 3 8B14.38

Interactive version: theaggregate.ai/benchmark?slug=adaplanbench · How It Works · Data refreshed daily, snapshot 2026-09-29.