Huatai Securities: Focus on AI Applications and Domestic Models for Investment
Huatai Securities has released a research report indicating a shift in the large language model (LLM) market competition. The focus is moving from capability rankings to cost-effectiveness for equivalent intelligence, driven by price reductions. OpenAI lowered prices for its Terra and Luna models by 20% and 80% respectively on July 30th. The following day, DeepSeek V4 Flash 0731 achieved a score of 50 on the Artificial Analysis Intelligence Index, just one point below Luna. Its hybrid price and average task cost are approximately $0.06 per million tokens and $0.03, representing significant cost reductions compared to Luna. Huatai Securities highlights that Kimi K3, with a score of 57, represents the high end of capability for current domestic open-weight models. DeepSeek V4 Flash has established a new cost benchmark within the 50-point capability range. Chinese models are now pursuing two competitive strategies: strong capabilities and high cost-effectiveness. This dual approach is expected to stimulate growth in downstream applications. Consequently, Huatai Securities recommends focusing on two main investment themes: AI applications and domestic LLM development.
AI model development is increasingly characterized by a price-performance race, moving beyond pure capability benchmarks. The significant price cuts by major players like OpenAI, alongside rapid advancements in cost-effective domestic models such as DeepSeek V4 Flash, suggest a market maturation. This dynamic implies that accessibility and affordability will become key differentiators, potentially accelerating the adoption of AI technologies across various industries. The emergence of distinct competitive strategies among Chinese models, balancing high performance with lower costs, presents an opportunity for market share gains. Investors are advised to consider the evolving landscape where cost efficiency may drive broader application deployment, rather than solely focusing on the highest theoretical capabilities. This trend could reshape the competitive advantages within the global AI ecosystem over the next decade.
AI-generated to prompt reflection — not editorial opinion, not advice, not a statement of fact. How this works.
