Cathay Pacific Haitong: Fable 5.2 released to promote the penetration of generic models into the strategic layer of embodying intelligent actions, supply chain companies are expected to benefit

The Zhitong Finance App learned that Cathay Pacific Haitong released a research report saying that Anthropic released Claude Fable 5.2/Mythos 5.2 to further enhance the robot motion decision-making capabilities of the GM large model and promote the penetration of the generic model into the embodying intelligent action strategy layer. Supply chain companies with core production links and layouts in North America are expected to benefit.

Cathay Pacific Haitong's main views are as follows:

Incident: Fable 5.2 is far ahead of Astra in public robot simulation operation tests, and has demonstrated the ability to operate a real machine.

On November 9, 2026, local time, Anthropic of the United States released Claude Fable 5.2/Mythos 5.2 to further enhance the robot motion decision-making capabilities of the GM large model and promote the penetration of the generic model into the artificial intelligent motion strategy layer. In the RoboDojo benchmark, Fable 5.2 scored 64.9%, about 2.2 times that of GPT6-Astra, an increase of 37.3 pct compared to the previous generation Fable 5.1, and 25.5 pct higher than Opus 5.5. This is important new evidence that GM's big model has entered the field of embodied intelligence. The bank believes that the general basic model of different companies is likely to cross the threshold of ability to make concrete decisions, and the pace of improvement is rapid. This will greatly speed up the process of commercializing the intelligent industry.

Previously, Astra has proven that the general-purpose model can directly output physical action instructions without special robot fine-tuning.

Fable 5.2 once again proved that the general basic model may gradually absorb the high-level action decision-making capabilities of dedicated VLAs, and the boundary between robot models and general models is becoming blurred. The independent value of dedicated VLAs faces reassessment. The better the general model is at directly outputting motion targets, the more the dedicated VLA needs to improve its advantages in terms of inference costs, real robot success rate, motion frequency, and deployability.

Generic models cannot replace real-time control systems for dedicated robots, and the value of JEU at the executive level may be even more prominent.

The large model is responsible for determining motion, and actual execution also requires stable state estimation, force control, collision response, joint control, and real-time end side closure. Increased general-purpose model capabilities may also increase the demand for high-quality execution systems.

Risk warning: The development of embedded intelligent model technology falls short of the expected risk, and downstream demand falls short of the expected risk.