Google has pushed the general release of Gemini 3.5 Pro to July, after the flagship model missed the June launch window the company set when it unveiled it at its I/O developer conference on May 19. As of late June the model remained in a limited Vertex AI enterprise preview rather than general availability.
The biggest context window yet
On paper, Gemini 3.5 Pro is ambitious. Google has said it carries a 2-million-token context window — double the one-million-token window on Gemini 3.5 Flash and, by Google's account, the largest of any production frontier model. It also adds a built-in Deep Think reasoning mode aimed at the hardest scientific, mathematical and coding problems.
Google has not yet published official benchmarks; the company said full results will arrive with general availability. Third-party estimates put pricing near $2 per million input tokens and $12 per million output tokens, though Google has not confirmed final rates.
Why Google is holding it back
The company has tied the delay to feedback from enterprise testers, citing excessive token consumption in extended, agentic tasks. Early testing also flagged coding performance that was not yet at the level Google expects from a flagship tier, and long, multi-step reasoning that fell short of the bar the company set for the model.
The slip has drawn attention beyond the product itself. According to reports, several senior Google researchers left for rival Anthropic around the same period, and Google's market value fell sharply on investor concern about the delay. Google has continued to ship elsewhere in the lineup, with Gemini 3.5 Flash reaching general availability at I/O.
A crowded release window
The timing is awkward. Anthropic shipped Claude Sonnet 5 on June 30, and OpenAI has been staging the release of its next model, leaving Gemini 3.5 Pro as one of the last flagship launches of the season still pending. A July release could also collide with a coming US framework on frontier-model releases that may classify the model as a "covered" system subject to government review before broad deployment.