RoboWits (VLM Planner) - Mutations: leaderboard
RoboWits (Lin et al., 2026) is a simulated bimanual manipulation benchmark of creative problem solving: seed tasks need geometry, material or assembly reasoning, such as using a rod to reach an object or displacing water to raise its level. Here a vision-language model plans in closed loop over perfect action primitives with oracle object states, on the mutations of 10 seed tasks that add a constraint breaking the usual solution; success rate in percent, higher is better.
Metric: Task success rate (%; modular planner with oracle object states and perfect action primitives, the listed vision-language model planning in closed loop; mean over the 10 evaluated seed tasks' mutations). Source: arxiv.org. Saturation forecast: Around February 2027. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 47.3 | #135 |
| 2 | GPT-4o | 44.4 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=robowits-vlm-planner-mutations · How It Works · Data refreshed daily, snapshot 2026-11-12.