The answer is obviously not enough.
In the recent competition of models and products, Google's pace has been a bit slow. This is true whether compared to the period after the release of Gemini-2.5 last year, or compared to recent moves by companies like Anthropic.
The rumored Gemini-3.1-Pro-Preview is about to be released. According to my speculation, it still won't be the strongest model for programming. In terms of model optimization, programming and other capabilities have increasingly shown conflict. The former can be continuously improved by allocating more "thinking budget" and large amounts of generated data, but these two factors affect the model's other abilities. We have too much knowledge and common sense judgment that is not derived from thinking and generated data.
Seedance 2.0 has surpassed Veo-3.1 in terms of video rationality. Although Veo-3.1 might still have slightly better color and lighting texture, in terms of practicality, Seedance 2.0 has pushed video generation models forward by a major level.
Claude Cowork is a very differentiated product. Much of our work requires tasks to be completed automatically by AI, not necessarily through AI programming. In more and more scenarios, it is impossible for us to orchestrate an Agent's workflow in advance; instead, we operate in a step-by-step manner. Cowork is suitable for this state.
In Gemini's multimodal map, there are at least TTS models and music generation models. Although there have been some minor updates, they haven't moved into the Gemini-3 era synchronously.
Also, don't forget that Google has an easily overlooked model line: the open-weight Gemma series.
Perhaps Google is facing a severe shortage of computing power. Yes, a massive user base isn't always a good thing because it means tight computing resources. When combined with a product line that is a bit "messy," overly dispersed resources might significantly affect the launch of new products and the improvement of user experience.
The things I mentioned above were originally expected to be updated within three to four months after the launch of Gemini-3. Now, nearly three months have passed. This can be understood as the time window opening, or it can be understood that Google is just slow.
Of course, what Google needs most is vigorous product integration: Imagine if NotebookLM were unbundled and could interconnect with more products, even AI Studio's Build or Antigravity and the Gemini CLI—how many possibilities would that create?
Of course, it's not just a matter of possibility; it's also that we now need an All-in-One application that is more versatile and productive than NotebookLM. It shouldn't just benchmark against Cowork, but be much more comprehensive, at least in multimodal capabilities.
Google is fully capable of doing this. If it can integrate Build, Stitch, Mixboard, Flow, or Whisk into one, and add support for Workspace and Google Drive, then it could be invincible.
Unfortunately, invincibility doesn't exist.