AsyncHow (Planner): leaderboard
AsyncHow asynchronous planning (Lin et al., 2024) as run in Robust Asynchronous Planning via Auto-Formalization (Zhang et al., 2026): 320 WikiHow-derived plans whose steps have durations and precedence constraints, and the answer is the shortest total time with parallel steps. Here the LLM writes the schedule directly, zero-shot at temperature 0; makespan accuracy in percent, higher is better.
Metric: Makespan accuracy (%; share of the 320 AsyncHow asynchronous WikiHow plans whose predicted total time equals the optimal makespan; the LLM writes the schedule directly (Planner); zero-shot, temperature 0). Source: arxiv.org. Saturation forecast: Estimated already saturated. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 96.56 | #349 |
| 2 | DeepSeek V4 Flash | 96.56 | #254 |
| 3 | Qwen 3.6 35B A3B | 95.63 | #461 |
| 4 | Gemini 3 Flash | 88.75 | #159 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=asynchow-planner · How It Works · Data refreshed daily, snapshot 2026-11-12.