Big model manufacturers are taking the lead alternately, each leading the way. In the fourth quarter of last year, it was Gemini, in the first quarter of this year, it was Cluade, and now it feels like it's GPT5.5's turn. The core is that everyone is actively striving for progress and iteration, and often leading is due to the opponent's laxity. Authropic, which was very restrained in its investment in computing power last year, should have felt the pain of computing power shortage recently. Sort out the timeline 1. Last year's fourth quarter: The Google Gemini 3 series (especially the Gemini 3 Pro/Deep Think around November 25) has made a strong breakthrough, leading many benchmarks such as Humanity's Last Exam, inference, and multimodality, triggering OpenAI's "Code Red". It is particularly outstanding in multimodal, long context, and search integration, and was once considered a turning point. 2. In the first quarter of this year: The most dazzling feature is the Anthropic Claude 4.6 (Opus/Sonnet, released in February), which has a strong comeback, especially leading in coding (SWE Bench), long context inference, agentic tasks, and actual productivity. 3. In April 2026: OpenAI GPT-5.5 is currently in the release window, just It will soon be fully launched on ChatGPT and Codex. It emphasizes better contextual understanding, coding, computer use, and agent capabilities, attempting to catch up and surpass. There is currently no absolute "king" in the field of big models, but rather each has their own expertise: 1) Claude: Focusing on coding, long tasks, and reliable reasoning often leads the way, with a lot of iterations on agents. 2) Gemini: Multi modal, fast, cost-effective, and strong over long text (3.1 Pro Preview still works well). 3) GPT stands out in terms of universal agents, tool invocation, real-time applications, and ecological integration, with significant improvements in context and specific professional tasks in the new version. The most active hoarding of computing power 4) Grok: There is X, a real-time content platform that provides a continuous stream of training data, but recently it has been constrained by team turbulence. Looking at the collaboration with cursor and the large amount of computing power accumulated, will we continue to keep up in the future. Of course, there is also Meta's latest Muse Spark, the latest work from the AI team. Not to mention that domestic brands such as Doubao, Qianwen, Hunyuan, and Kimi also have their own unique characteristics and are killing crazy. This is the current situation in the field of big models: rapid iteration and alternating dominance. There is currently no company that can lead or even monopolize in the long run. Of course, competition is a good thing for users - as models become stronger and prices/speeds are optimized. In mid February, I talked about it in a tweet titled 'The War on Capital Expenditure': 'The more you go downstream, the more intense the competition becomes on the user side, and of course, it is also the key to the future victory of AI.'. It can be said that big models, agents or applications targeting B-end or C-end are the crown of AI, but in the short term, it is difficult to see who will become the true winner at this level, and many times it is alternating to take the lead. Now it seems that way After the release of GPT-5.5, it is estimated that there will be another round of chart topping. Who will take the lead next?