Google’s next flagship artificial intelligence model might be dropping much sooner than everyone thought. Recent leaks show that the tech company is getting ready to launch the Gemini 3.5 Pro model with major upgrades designed to challenge rival products like Claude and OpenAI.
A massive token window changes how data is processed
One of the biggest rumors surrounding the upcoming release is a 2-million-token context window. This represents the largest production context available from any major provider right now. To put that massive number into perspective, a two-million token limit is large enough to swallow an entire codebase, a full year of customer support records, or tens of thousands of rows of financial data, all in a single prompt.
This kind of capacity means users will not have to chunk large documents anymore. It is specifically built for tasks that need heavy context, like whole-repository security audits or cross-document legal analysis where connections matter more than speed.
Deep Think mode tackles multi-step problems with accuracy
Another major leaked feature is the new Deep Thinking mode. The Google team designed this specific tool for complex problem-solving rather than rapid responses. The model actually spends more computing power reasoning through a problem before it answers.
This feature will be heavily optimized to handle long-running agent workflows and deep research projects. By trading instant latency for a deeper thought process, the system can provide reliable answers for difficult tasks like mathematical proofs or policy analysis instead of just rushing to complete the prompt.
Major improvements target code generation and web interfaces
The leaked reports also claim that the Gemini 3.5 Pro variant brings massive leaps in software development. Testers report that early builds are capable of writing clean and complex code on the first attempt. The upcoming Pro version will also lock into Google’s native computer use upgrade, allowing AI agents to interact directly with desktop and web user interfaces.
As of late May 2026, the company opened limited access to Vertex enterprise customers, and general availability is expected in late June. If the benchmarks hold up, this new release could easily mark the most disruptive and capable product drop of the entire year.
