Anthropic: LLMs fail at direct robot joint control but succeed supervising controllers
From Anthropic’s ‘Claude plays robotics’ study.
Confirmed
The takeaway
On July 9, 2026 Anthropic published “Claude plays robotics”, which tested twelve models from five providers on simulated and physical robots.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 2 of 5
- Last verified
- 1 October 2026
Your AI and this story
- GPT-6.2 Astra39 days after its cutoff
- Claude Opus 5.59 days after its cutoff
- Gemini 4 Hydrogen100 days after its cutoff
- Grok 4.739 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 9 days before it.
Key facts
- Authors: Shmuel Berman, Michael Ilie, Jia Deng, Daniel Freeman
- 12 models from 5 providers, several Claude generations
- Direct manipulation success 0–5.5%; Claude Mythos Preview highest at 5.5%
- Quadruped: newer models reach ‘limited but meaningful’ whole-body control (balancing almost two seconds); no model stood a collapsed humanoid up
- VLA supervision on LIBERO-40 raised success substantially for all models
- A compass tool beat visual self-views; extended reasoning helped little and sometimes hurt
What happened
A systematic eval of general LLMs as robot controllers at several levels of abstraction, from raw joint commands up to supervising pretrained policies.
Why it matters
The study measures how far general frontier models are from embodied control. The answer depends heavily on the interface: high-level supervision already works, while low-level control improves only now and then from one generation to the next.
Sources
1 source from 1 site. Numbers match the chips in the text.
1 source: 1 primary
Primary
- Anthropic: Claude plays roboticsanthropic.com, official
Changes
- Filed