Anthropic releases Claude Fable 5.2, Nvidia releases GR00T 2, humanoid robot concept strengthens
East Money reported that on November 9, Anthropic released its new-generation flagship model Claude Fable 5.2, with upgrades in complex programming, scientific research reasoning, chart understanding and robot control, achieving a 64.9% success rate in the RoboDojo simulated dual-arm robot test, higher than GPT-6 Astra's 29.8%. On the same day, Nvidia released the world action model GR00T 2, whose pretraining integrated 403,000 hours of human first-person video, 10,000 hours of UMI gripper data and 23,500 hours of robot teleoperation data; after adding complete human video, the average completion progress on 72 real-machine tasks rose from 50% to 70%. Affected by this, the humanoid robot concept strengthened on November 10, with Yijiahe hitting the limit up and Unitree Technology rising nearly 9%.
Why read this
AI score 52
The article puts together the robot capability progress of Claude Fable 5.2 and Nvidia GR00T 2 with brokerages' judgments on the physical AI route, making it easier to understand industry expectations.
Topics
Body · AI translation
Article body EM_StockImg_Start
EM_StockImg_End
On Tuesday (November 10), the humanoid robot concept gained strength. As of the midday close, Yijiahe hit the daily limit, and leading player Unitree Technology rose nearly 9%, with cumulative gains exceeding 90% since October 16; Jingye Intelligent, Leadshine Technology, Sanhua Intelligent Controls, Leaderdrive and others followed the rally.
Major moves by Anthropic and Nvidia
On the news front, on November 9 local time, Anthropic released its new-generation flagship model Claude Fable 5.2, upgrading capabilities in complex programming, scientific research reasoning, chart understanding and robot control. Fable 5.2 achieved a 64.9% success rate in the RoboDojo simulated dual-arm robot test, more than double that of GPT-6 Astra (29.8%), and also demonstrated footage of Fable 5.2 controlling robots to complete tasks such as unscrewing bottle caps, manipulating a Rubik's cube and drawing.
On the same day, Nvidia released the world action model GR00T 2, which jointly predicts future frames and actions, and improves robot generalization and fine manipulation capabilities through pretraining on massive human experience and a small amount of real-machine teaching. The pretraining integrated 403,000 hours of first-person human video, 10,000 hours of UMI gripper data and 23,500 hours of robot teleoperation data. After adding complete human video, the average completion progress on 72 real-machine tasks rose from 50% to 70%.
Huayuan Securities believes that GR00T 2 further validates the path of "pretraining on massive human experience + post-training with a small amount of real-machine data," with model generalization and complex manipulation capabilities continuing to improve. The Tianji robotic arm and Sharpa dexterous hand participated in core validation, highlighting the value of collaboration between dexterous arms plus dexterous hands and the model.
Institutions show consistent expectations on the direction (the inflection point of physical AI) and the convergence point (whichever route wins, the hardware interface layer is a common need): Guojin Securities pointed out that a mature embodied brain requires tens of millions or even hundreds of millions of hours of data, while the industry currently has only hundreds of thousands of hours, a supply-demand gap of 2-3 orders of magnitude, and data collection is the segment with the highest order visibility.
Huatai Securities believes opportunities are concentrated in motor, lead screw, sensor and reducer companies with real orders; GF Securities believes the industry focus has shifted from body manufacturing to embodied models.
Main funds accumulate multiple stocks
East Money Choice data shows that since early October this year, main funds have accumulated a batch of robot concept stocks against the trend.
Among them, Rockchip ranked first, with net main-force buying of nearly 1.1 billion yuan; Unitree Technology ranked second, with net main-force buying of nearly 900 million yuan.
Moore Threads-U, Molding Technology, Wanxiang Qianchao, Ningbo Dongli, Everwin Precision, Ampron, Guangyang, Zhejiang Rongtai,
For stocks such as Aolaide, SW New Materials, and Hanwei Technology, net main-force buying ranged from 870 million yuan to 300 million yuan.
Institutions: Complete machines and components are expected to benefit
On the Tesla side, Musk recently said that "Optimus" will begin deliveries around the end of 2027, but the delivery format may be "rental" rather than "sale." Musk said the initial cost of Optimus may be similar to that of an electric vehicle, with the initial cost per robot at about $30,000, later falling to about $20,000. Musk expects Optimus to take at least a year to reach full mass production, and before external deliveries, Optimus will first be used in Tesla's own factories.
Sealand Securities believes that Tesla's humanoid robot has entered production, with capacity expected to reach 20,000 units per week before the end of 2027, and it plans to deliver to external customers, while achieving 400 megawatts of AI training compute by the end of the year to support robot intelligence, bringing deterministic growth opportunities to the industrial chain. In addition, the capital side has significantly recovered, with the robot industry index ranking among the top in net inflows, showing increased market allocation willingness.
Orient Securities issued a research report stating that, combined with the performance of general large models such as OpenAI GPT-6 Astra and Google Gemini 4 Argon in the field of zero-shot generalization for robots, it is expected that embodied robot models are moving from task specialization toward highly generalized general models. The leap in embodied model capabilities will also accelerate the robot industry's advance from prototype demonstrations to practical scenario applications. Data collection capabilities and large-scale robot mass production capabilities will become key competitive links in the future robot industry, and companies related to complete robots and core components are all expected to benefit.
From an investment strategy perspective, the robot, data center, liquid cooling, and gas power generation industrial chains will usher in catalysts, and it is recommended to continue paying attention to targets in the liquid cooling industrial chain, the gas generator and diesel generator industrial chains, and the humanoid robot industrial chain.
(Article source: East Money Research Center) Middle of the main text Embedded advertisement