SemiAnalysis is already predicting that Meta will surpass Google within six months. As mentioned in this tweet from July 4th: 'The performance of next-gen models is the key to winning. Everyone must immediately surpass GPT-5.5 and Claude Fable 5. As long as Meta's next-gen model performance can approach the latest generation and exceed the previous generation of O/A/G, I personally think they can still secure a spot. If they can catch up, I believe Meta has a big chance.' Looking at Muse Spark 1.1 now, this seems to be the case. While there’s still a gap compared to the latest generation from Anthropic and OpenAI, its performance has already surpassed the previous generation models from A and O. As long as they can deliver competitive performance, the entire competitive landscape could shift immediately. Especially with Muse Spark 1.1’s extremely aggressive pricing strategy—about one-fourth the price of OpenAI and Anthropic’s top-tier models. It’s clear that Zuckerberg’s approach to large models is actually similar to Microsoft’s strategy with MAI models: no longer aiming to completely outperform the latest generation models from Anthropic and OpenAI in terms of performance, but instead focusing on being 'good enough' with highly competitive pricing. This 'good enough' performance (surpassing the previous generation models from O and A) combined with cost-effective token pricing could capture a significant share in the future, where enterprises adopt AI for specific scenarios and multi-model architectures. This shift in enterprise AI adoption was discussed earlier here: https://(((x.com)))/qinbafrank/status/2074754779755295164?s=46&t=k6rimWsEbo2D2tXolYcM-A. SemiAnalysis believes that Meta’s ability to surpass Google lies in three core advantages: data, talent, and in-house compute power. This aligns with what I mentioned earlier here: https://(((x.com)))/qinbafrank/status/2072902811675922742?s=46&t=k6rimWsEbo2D2tXolYcM-A. What truly sets Meta apart for the future is its in-house models + proprietary user scenarios + closed-loop advertising/social data + in-house inference infrastructure. I made this prediction a bit earlier than SA did . That said, it’s still too early to declare Google as definitively falling behind. Google has its own strengths, as discussed earlier here: https://(((x.com)))/qinbafrank/status/2047491554609422452?s=46&t=k6rimWsEbo2D2tXolYcM-A. The competitive landscape for large models is one of alternating leadership, with each player shining for a few months at a time.