Google Unveils Gemini 3.5 Pro Tomorrow, but Keeps the Details

Google could unveil a new artificial intelligence model, Gemini 3.5 Pro, as soon as July 17. The date has been cited by several technology outlets, but the company itself has not officially confirmed it, nor has it disclosed the model’s specifications. Everything known so far comes from leaks and media reports, not from company statements.

That caveat matters. In a race where announcement dates are used as a weapon against competitors, unofficial information spreads quickly and often fails to match what is finally released. So what follows should be read as what is being reported, not as what has been confirmed.

What is known, and what is only assumed

Two claims are drawing the most attention. The first is a context window of two million tokens, the amount of text a model can keep in view during a single conversation. If accurate, it would mean Gemini 3.5 Pro could process very long documents, entire books or large codebases, without losing the thread halfway through.

The second is a layer called Deep Think, described as a mechanism for multi-step reasoning. The idea is that the model does not answer immediately, but breaks a problem into steps before drawing a conclusion. A similar approach already exists in competing models and tends to help with tasks in mathematics, logic and programming.

The most important point for this moment bears repeating: none of these figures has been confirmed by an official source. Preliminary specifications circulating online have proven both accurate and wrong in the past. Until an official release, the two-million-token context and Deep Think remain expectations, not facts.

Why did Google reportedly start over?

Part of the reporting also addresses why the model is late. According to those accounts, Google scrapped its original base model and started again from scratch. The reasons cited are problems with recursive tool calling, where a model loops through its own functions, and with generating SVG graphics.

If true, such a decision says more about the process than about the product itself. Restarting midway through development is expensive and rare, and usually means a team judged that the underlying problem could not be patched, only solved with a new foundation. It is worth repeating that this too is unconfirmed, a reconstruction from several sources.

The context around the announcement is clear even without leaks. China is hosting a summit on artificial intelligence these days, and China’s DeepSeek is preparing a model of its own. The pace of releases among the major players keeps accelerating, so launch dates are read as part of the competition, not merely as a technical moment.

For now, then, we have an expected date, a handful of reported specifications and one story about a fresh start, all without official confirmation. Which of it holds up we will know only once Google publishes its documentation. Until then, the fairest thing to say is that we do not know how different Gemini 3.5 Pro will really be from its predecessor.

With announcements like this, are you more interested in the number in the specifications, or in what the model actually does once you try it?