Anthropic and NVIDIA unveiled their new offerings on the same day. Are robots truly on the verge of "growing their own brains"?

① As a general large language model, Anthropic's Fable 5.2 has demonstrated outstanding capabilities in robot control.

② NVIDIA has launched the new-generation robot model GR00T 2, and released a robot learning framework and a number of related technologies.

③ Brokerage firms point out that the entry of large model manufacturers will accelerate the progress of embodied models, and the commercialization expectation of the industry is continuously rising.

On November 9 local time, NVIDIA and Anthropic successively released new products, adding two major boosts to the robotics sector.

First, Anthropic, which was under intense scrutiny from the capital market on the eve of its IPO, unveiled two heavyweight new models: the new-generation flagship models Fable 5.2 and Mythos 5.2.

Among them, as a general large model, the capabilities demonstrated by Fable 5.2 in robot control have attracted widespread attention in the industry. The test data released by Anthropic shows that in the RoboDojo dual-arm simulated robot benchmark test, the average success rate of Fable 5.2 reaches 64.9%.

What does a 64.9% success rate mean? The previous generation large model Fable 5.1 only had a 27.6% success rate in this test. In other words, the success rate increased by 37.3 percentage points after just a "0.1 generation" update; in the same comparison of Fable 5.2 this time, Claude Opus 5.5 scored 39.4%, while GPT-6 Astra scored 29.8%.

A higher success rate score also means that the model has a stronger ability to judge the environment and arrange actions based on camera footage.

Moreover, Fable 5.2 also has generalization capabilities. Anthropic pointed out that "Fable 5.2 demonstrates greatly improved capabilities in controlling physical robots, even for tasks that have never been explicitly trained." In the public demonstration video released by Anthropic, Fable 5.2 controlled a physical robotic arm to complete three specially untrained tasks: drawing patterns, screwing bottle caps, and solving Rubik's cubes.

Anthropic stated that Fable 5.2's performance in robot control is "better than most AI models specifically designed for robot control."

This is also an important reason why Fable 5.2 has caused a huge response - before this, the mainstream view was that robots needed more vertical large models specifically for embodied intelligence to effectively solve spatial understanding, action arrangement and control problems. As a general large model, Fable 5.2 not only has outstanding performance in complex programming, scientific research tasks, computer operation, interdisciplinary reasoning and chart recognition, but also achieves a huge leap in robot control.

On the other side, on the same day of November 9 local time, NVIDIA announced the launch of the brand-new new-generation robot model GR00T 2, and released the robot learning framework Isaac Lab 3.0 and a number of robot training and evaluation technologies.

NVIDIA said that GR00T 2 is an open cutting-edge world model built on the Cosmos platform. The model is trained with robot and human data, including 100,000 hours of open-source human first-person perspective data provided by NVIDIA.

One of the main features of GR00T 2 is its out-of-the-box task execution capability: among 50 tasks, 40 of them can be completed directly without special demonstration training for specific tasks. In addition, the model also supports robots to master skills by learning human operation demonstrations, and can transfer the learned skills to robots of different forms without redoing task demonstrations for each robot form. According to NVIDIA, with only 1 demonstration provided for each task, the average success rate of GR00T 2 in 50 tasks reaches 71%; if 5 demonstrations are provided for each task, the average success rate rises to 95%.

Earlier on November 5, NVIDIA just released 4 Cosmos 3 Nano open-weight models, covering multi-view scene generation, interactive simulation and robot motion prediction; the Apple research team released the HALO humanoid robot autonomous control system on November 4, which realizes the robot's autonomous execution of tasks such as walking, obstacle avoidance, sitting down and pushing boxes in unmapped real environments by synthesizing training data in 1000 programmatically generated home scenes.

It is worth mentioning that on November 4, Wang Xingxing, Founder, Chairman and CEO of Unitree Robotics, said that 2027 will most likely become the critical point for the development of the robotics industry or robot models, and robot capabilities are expected to be greatly improved next year; the "ChatGPT moment" of embodied intelligence is just around the corner. The model is exactly the technical threshold to realize this moment: AI models do not yet have a sufficient understanding of physical contact and tactile sensation, and there is a deviation between the input and output of the model and the alignment of the real world. He predicts that this problem is expected to be solved in the next one to three years.

The research report of Orient Securities on November 2 pointed out that the main reason for the previous market's pessimism about the embodied intelligence industry is that people believed that embodied models are different from large models, and worried that the slow progress of models of embodied industry manufacturers would lead to limited application scenarios of embodied intelligence. However, as GPT-6 Astra has achieved good practical application in the embodied field, some investors gradually begin to believe that general large models are also expected to seize market opportunities in the embodied field. The entry of large model manufacturers will accelerate the progress of embodied models, so the commercialization expectation of the industry is continuously rising.

Analysts further pointed out that large models have huge potential as the strategy layer of embodied intelligence, and the manufacturers of embodied bodies benefit the most marginally. With the continuous exploration of general large model manufacturers in the embodied intelligence field in the future, it is expected to drive the accelerated application of embodied intelligence in the fields of logistics and industrial manufacturing, and the barriers of manufacturers with vertical scenario know-how will be relatively higher. It is expected that more large model manufacturers will accelerate their layout in the embodied field in the future, which is expected to promote breakthroughs in more application scenarios, thus driving the accelerated breakthrough of the industry's commercialization process, and robot complete machine and component manufacturers are expected to benefit.

This article is from "CLS", authorized for release by 36Kr.