Anthropic Unveils Claude Fable 5.2 with Enhanced Robotic Control; Former Tesla Optimus Researcher Joins

Machine Mind Editorial Department

Early this morning, Anthropic released two top-tier models, Claude Fable 5.2 and Mythos 5.2, in one go. The new models demonstrated several achievements in scientific research, including the formalization of the proof of Fermat's Last Theorem, completed in only 30 hours through the collaboration of about 1000 AI agents.

In this update, robot control capabilities also received significant attention. In the official release article, the Science and Engineering section specifically set up a “Hardware and Robotics” subsection.

In the official demonstration video, Claude Fable 5.2 controlled a robot to complete tasks such as manipulating puzzle blocks, unscrewing a bottle cap, and drawing a robot pattern.

According to data released by Anthropic, Claude Fable 5.2 reached a score of 64.9% on RoboDojo Sim, an increase of 37.3 percentage points over the previous generation Fable 5.1's 27.6%, and it also exceeded Claude Opus 5.5 (39.4%) and GPT-6 Astra (29.8%).

Anthropic also stated that Fable 5.2 outperformed most AI systems specifically developed for robot control in this evaluation.

Anthropic Product Manager Alex Albert stated bluntly that the “bitter lesson” of robotics has arrived.

At the same time, Bingjie Tang, a researcher who previously participated in the development of Tesla's Optimus humanoid robot, has also joined Anthropic. She previously mainly worked on robot dexterous manipulation, policy learning, and real-world deployment, and her research background forms a noteworthy intersection with the robot capabilities Anthropic recently demonstrated.

Anthropic is expanding Claude's robot control capabilities

In fact, Anthropic has publicly released robot-related research multiple times this year.

In July, Anthropic released “Claude plays robotics,” testing Claude's performance on simulated robots, robotic arms, and real quadruped robots.

Researchers tried various control methods, including generating low-level control instructions, writing control programs, and calling pre-trained robot policies.

Experiments found that Claude already possesses a certain level of robot environment understanding and operation planning capability, with specific performance being influenced by factors such as control interfaces, action precision, and inference latency.

By October, research institutions such as Robocurve published the paper “The Bitter Lesson Comes to Robotics,” further evaluating the ability of frontier general multimodal models to control real robotic arms.

The study found that general models are already capable of completing various physical robot operation tasks, and relevant capabilities have continued to improve over several months. At the same time, operation precision, execution speed, and inference costs remain important constraints in practical applications.

One of the paper's authors, Jay Chooi, also mentioned the “bitter lessons” of the robotics field on social media after the release of Fable 5.2.

“Bitter lessons” originates from a classic viewpoint in the field of machine learning: as computing resources and model scales grow, general learning methods often achieve competitiveness on more and more tasks.

Now, this discussion has also begun to extend to the field of robotics.

As the visual understanding, spatial reasoning, and task planning capabilities of general models continue to improve, researchers are exploring to what extent they can perform work in robot control systems.

Former Tesla Optimus researcher joins Anthropic

Just as Anthropic demonstrated robotic capabilities, a piece of news regarding talent flow also drew attention.

Researcher Bingjie Tang, who was previously involved in the development of Tesla's Optimus humanoid robot, recently confirmed joining Anthropic as a Member of Technical Staff.

Personal homepage: https://bingjietang718.github.io/

Previously, she worked on learning-based dexterous manipulation research in the Tesla Optimus AI team, with work covering data collection, robot policy training, evaluation, and real robot deployment.

Before joining Tesla, Bingjie Tang obtained a PhD in Computer Science from the University of Southern California. During her PhD, she collaborated closely with researchers from NVIDIA Seattle Robotics Lab to conduct research related to contact-rich robot manipulation, robot learning, and Sim-to-Real.

Among them, Sim-to-Real is one of the important directions of current robot learning, mainly studying how to transfer operation policies trained in a simulation environment to real robots.

Friction, contact status, object shapes, and hardware errors in real environments will all affect the robot's action performance; therefore, from simulation training to real deployment, there are still a large number of technical problems that need to be solved.

The Refinery and FORGE research that Bingjie Tang previously participated in also involve robot policy learning, real interaction, and manipulation adaptation.

From her personal resume, she has research experience covering data, algorithm training, to real robot deployment, which has a strong technical correlation with the robot capabilities recently demonstrated by Anthropic.

In 2026, embodied intelligence continues to heat up. From world models and spatial intelligence to robot foundation models, more and more AI companies are beginning to explore the physical world, while the robot control capabilities of general large models are also rapidly progressing.

As the capabilities of general large models continue to improve, what roles will they play in future robot intelligent systems? What new progress will Anthropic bring next? It is worth looking forward to.

© THE END

For republication, please contact this official account for authorization

For submissions or media inquiries: [email protected]