Does Meta have no chance at all on the big model? Personally, I don't think so. A few points: 1. Even if Meta takes the closed source route, it must be acknowledged that OpenAI and Authropic are far ahead, but it cannot be said that Meta has no advantage. There are still moats that OpenAI and Anthropic cannot compare to: 1) Deep product integration: Meta AI is directly embedded into WhatsApp, Instagram, Facebook, Messenger, etc., covering billions of users, and can achieve personalization (based on social data), multimodal (visual+chat), and advertising conversion optimization. 2) Infrastructure and cost advantages: Meta is building a giant computing power cluster with a huge scale of its own data center, which has the potential to lower inference costs and latency. In fact, looking at how Google caught up last year, on the one hand, Gemini3 had strong performance, and on the other hand, it was also a powerful distribution channel for Google. At present, Meta's problems have not yet reached the forefront of large-scale models. If the performance of the next generation of large-scale models can be improved, coupled with Meta's powerful distribution channels, at least they can have a place on the C-end. Moreover, the effectiveness of AI in advertising is very direct. Previously, Meta's own model collapsed and borrowed the API from Gemini. However, due to excessive computing power requests, Google restricted access. So the core is not whether Meta has a chance in AI, but whether Meta's next-generation models can catch up in performance. If we can catch up, I think Meta has a great chance. The model has good performance, strong distribution channels, and the user base can naturally increase, and the promotion of agents can also be accelerated. Then the B-end programming is weak, but the B-end AI advertising should be able to eat a lot. 2. Speaking of open source, Meta has not given up on open source either. Currently, Meta should adopt a mixed strategy. Muse Spark is closed source, but the official statement is that they "hope for future versions of the open-source model" and are developing the next generation model for the open weight version (retaining some proprietary capabilities such as network security code generation) GLM is indeed very strong, and China's open source ecosystem for large models is also developing rapidly, with a momentum like a rainbow. As mentioned in the tweet, there are many regulations in Europe, America, and Japan that prohibit the use of Chinese models, and Meta could have filled in this non Chinese open source Tier-1 ecosystem. Meta has already entered through Llama and there is still room for improvement (continuing to open source some models+emphasizing 'Western trusted sources') What is the core discussed above? It's still the performance of the model, this is the winner. It is necessary to immediately surpass GPT 5.5 and Claude Fable5. As long as the performance of the next generation meta model can approach the latest generation and surpass the performance of the previous generation O/A/G, I personally believe that it can still occupy a place. But if the model performance doesn't improve consistently, it's really difficult to turn things around. Like yesterday here https://(x.com)/qinbufark/status/2072902811675922742? As discussed by s=46&t=k6rimWs Ebo2D2TXolYcM-A, what Meta will truly value and differentiate in the future is its "proprietary model+proprietary user scenario+proprietary advertising/social data loop+proprietary inference infrastructure" At present, the key point of Meta is the performance of the model. If it can be broken through, the future winner will be the Agent. Of course, if the future model really doesn't work, Meta will focus on being an agent, just like Tencent's current approach. Whoever has good computing power and high token cost-effectiveness can use it, and there are also opportunities. Meta also has advantages in pushing agents Simply put, Meta still has cards to play, but the AI field is changing rapidly, and the situation may be different in a few months. Just like before here https://(x.com)/qinbufark/status/2047491554609422452? S=46&t=k6rimWs Ebo2D2TXolYcM-A, we talked about leading the way in March and May.