At the beginning of the year, we said that major players in the large model space would each have their moment of glory for three to five months, but none would maintain long-term dominance or monopolize the field—it’s all about rapid iteration and taking turns at the top. Sure enough, after a few months of delay, Gemini 4 has finally launched. The community blind test results for Gemini 4 Argon (High) are impressive, especially in text synthesis capabilities, where it’s taken the crown. But this year, the market has already been bombarded by GPT 5.5, 5.6, 6 Astra, Claude Fable 5.1, Opus 5.5, and Grok 4—not to mention the even higher-frequency iterations of Chinese large models—it’s honestly starting to feel a bit numbing. Plus, it seems like the main battlefield for AI has shifted from the models themselves to Agents. Grok Bot, Meta Muse, OpenAI Dots, and others have been rolling out one after another recently, driving rapid iteration in this area. Google actually has a strong advantage in the Agent space too: massive user base, huge data accumulation, and powerful distribution capabilities. If they focus on this, it’s definitely worth keeping an eye on. Back in February, I mentioned this in my tweet "The War of Capital Expenditure": "Looking downstream, the closer you get to the user end, the fiercer the competition becomes. This is, of course, the key to determining the future of AI. You could say that large models, and the Agents or applications targeting B-end or C-end users, are the crown jewels of AI. But in the short term, it’s hard to predict who will ultimately emerge as the true winner—most of the time, it’s just alternating leadership." https://(x.com)/qinbafrank/status/2023597321368547520?s=46&t=k6rimWsEbo2D2tXolYcM-A