Gemini 3.5 Pro Is Finally Coming (Targeting July 17): 2M-Token Context and “Deep Think”

Google’s most-delayed model of the year is finally near the finish line. After being announced at I/O in May and then repeatedly slipping its general-availability date, Gemini 3.5 Pro is now reportedly targeting July 17 — though a quick, important caveat: Google hasn’t officially confirmed that date with a model card, API docs or pricing as of early July, so treat it as a strong target, not a promise.

The headline feature: a 2-million-token context window

The standout spec is a two-million-token context window — double Gemini 3.5 Flash and the largest deployed in any production frontier model to date. On paper, that means you can dump entire codebases, book-length documents, or sprawling data sets into a single prompt. The honest question, though, is whether the model can actually reason across all 2M tokens or merely retrieve from them — and that only gets answered by independent long-context evals, which don’t exist yet. Big window, unproven depth.

The AI tool stack actually worth paying for

One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.

Get the free stack →

Deep Think, gated behind the $250 tier

The other headline is Deep Think, an extended-reasoning mode aimed at hard, multi-step problems that benefit from the model “reflecting” longer before answering — the same territory as GPT-5.6 Sol’s ultra mode and Claude’s extended thinking. The catch: it’s reportedly locked to Google’s $250-per-month Ultra subscription. Frontier reasoning is increasingly a luxury SKU, not a default.

The takeaway

Gemini 3.5 Pro arriving — if July 17 holds — reloads the top of the model race just as GPT-5.6 goes public and Claude’s Mythos-class flagship is back online. That’s three frontier labs all shipping within weeks of each other, which is great for anyone buying. Just don’t pre-order the hype: a record context window and a premium reasoning mode are promising on the spec sheet, and completely unproven until the independent benchmarks land. We’ll update when it actually ships.

Sources

Scroll to Top