CICC: The open source model is improving at an accelerated pace, and the closed-source model is optimistic that the closed source model will jointly promote the implementation of AGI

The Zhitong Finance App learned that since this year, CICC released a research report saying that since this year, the level of intelligence of the open source model has accelerated, and its ability is approaching that of the top closed source model. The market focus has also extended from the hardware+closed source model to open source models, cloud vendors, application vendors, etc. The bank is optimistic about the open source model and the closed source model jointly promoting the implementation of AGI, recommending domestic open source model vendors; optimistic about the revaluation of cloud vendors supported by the open source model and recommending cloud vendors; optimistic about the accelerated development of FDE and AI applications under the penetration of the open source model, and recommended application vendors.

CICC's main views are as follows:

The open source model is trending up in token calls, and the market space is broad

The open source model is based on high cost performance and high intelligence, and has been widely recognized by users at home and abroad, and the number of open source model calls is trending up. The bank estimates that the revenue from overseas open source models will reach 10 billion US dollars this year, and in 2027, global open source model calls will account for 50%-70% of the total number of tokens, and the value will reach 12%-30% of the large model market. The bank believes that the accelerated penetration of the open source model may drive the overall token unit price downward, but based on its openness and inclusiveness, it will unlock more usage scenarios, and overall token usage may increase dramatically.

The open source model is based on an open strategy to improve the level of intelligence and the speed of commercialization

The open source model follows the principle of model primacy, adheres to the Scaling Law and increases training and reinforcement learning to continuously improve the level of model intelligence. Kimi K3 and Qwen 3.8 all reached a parameter scale of nearly 3T. The openness of the open source model enables it to implement more high-value application scenarios, and the increased level of intelligence also allows open source model manufacturers to obtain more bargaining power. For example, commercialization continues to accelerate through price increases or discussions and division with platform deployers.

Open source models promote ecological prosperity and drive increased demand from cloud vendors, FDE, and application vendors

There are many participants in the overall ecosystem of the open source model. Computing hardware, cloud vendors, inference and tuning platforms, and aggregation/application platforms will all participate. Technological advances in the open source model will also drive commercialization of relevant participants. Among them, the value increase brought about by cloud vendors deploying open source models, and FDE also plays an important role in deploying open source models and understanding enterprise workflow requirements. Application vendors improve innovation capabilities and penetration efficiency under the impetus of the open source model. In the future, as application-side TAM expands, it will also feed back the development of the open source/closed source model industry chain.

risk

AI technology progress falls short of expectations; downstream demand falls short of expectations; competition among manufacturers intensifies.