Huatai Securities: Computing power and model demand are expected to further spread from the training side to the high-intensity inference side.

Huatai Securities pointed out that on June 9, Anthropic released Claude Fable 5, with the following key highlights: Significant improvement in artificial intelligence level, with Fable 5 currently ranking first in the Artificial Analysis Intelligent Index; Continuous strengthening of agent capabilities, with a noticeable improvement in deliverability in long, multi-step, and complex tasks for Fable 5; Enhancement of visual recognition and long-term memory capabilities. For safety reasons, Fable 5 has set up barriers for network security, biochemistry, and model distillation scenarios, and related requests may trigger the model to revert to Opus 4.8. Huatai Securities believes that Fable 5 is expected to further expand the cutting-edge capabilities of long-range agents, promote the penetration of office agents, accelerate their landing in high-value projects in the B-end scenario, and ultimately spread to the domestic market. However, in the short term, frequent rollbacks due to safety interceptions may have a certain impact on the user experience for general consumers. With the accelerated commercialization of complex coding, office agents, and multimodal tasks, computing power and model requirements are expected to further spread from the training side to the high-intensity inference side.