Context windows have officially entered their "who can remember the most" arms race, and Google just showed up with an elephant. Gemini 3.5 Pro is now generally available, and it can hold two million tokens in its head at once — enough to swallow a mid-sized codebase and a stack of research papers without breaking a sweat.

What Actually Shipped

Google DeepMind announced general availability of Gemini 3.5 Pro on July 17, after reports the launch had slipped from an earlier June target while the team reworked coding and long-horizon reasoning performance. The headline specs are a 2-million-token context window — double what any other frontier model currently ships in production — plus a "Deep Think" reasoning mode built for complex, multi-step problems.

The model is rolling out through Google AI Studio, the Gemini API, and Vertex AI, sitting alongside the existing Gemini 3.x lineup rather than replacing it outright. Official pricing hasn't been published yet, though enterprise preview reports point to long-context surcharges once requests blow past 200K tokens.

The Context Window Flex Nobody Needed, Except Everyone

A 2-million-token window sounds like a spec-sheet stunt until you remember what people actually do with these things: dump in an entire repo, a legal discovery pile, or hours of video transcript and ask for one coherent answer instead of forty summaries stitched together with hope. That's a genuinely useful unlock, not just a bigger number for its own sake.

The quieter story is the delay itself. Google reportedly held the broader release back because internal testing showed the model falling short on coding and complex reasoning — a rare admission that "ship it anyway" isn't the plan, even with rivals like Anthropic and OpenAI breathing down its neck.

A bigger context window won't fix a model that can't reason its way out of a wet paper bag, but pairing the two is exactly the bet Google just made — a few months later than planned, but made all the same.

Source: Tech Times