Robot arm tests rank GPT-6 variants: better reasoning means better manipulation at 30x the cost

YuXiang_IRVL · x · 2026-09-23

In LIBERO manipulation tasks, GPT-6 Astra > Sol > Luna: stronger reasoning translates directly into better robot manipulation. But cost differs sharply — Astra spent $3.20 over 29 calls ($0.11/call) vs Luna's $0.23 over 62 calls ($0.0037/call), though Luna failed the task.

Related event: LIBERO Benchmarks Show Stronger GPT-6 Reasoning Yields Better Robot Manipulation(2 posts)→

Original post →

More from Embodied

Embodied channel →