Google is rolling out Gemini 3.5 Pro in July 2026 after pushing the model’s general availability from June to refine its performance. The frontier model represents a major step forward in reasoning, coding, and agent capabilities, doubling the context window of competing models to 2 million tokens.

The delay from June to July reflects Google’s commitment to quality over schedule. The company had unveiled Gemini 3.5 Pro at its I/O developer conference in May, but beta testers and early enterprise customers uncovered enough refinements needed that Google decided to postpone the public launch. That caution paid off—the model now sits in limited Vertex AI preview and is approaching general availability this month.

What Gemini 3.5 Pro Does

Gemini 3.5 Pro offers a 2 million-token context window, effectively doubling Claude Opus 4.8. For developers working with large codebases, long documents, or complex workflows, that context matters. The model also includes Deep Think reasoning, Google’s answer to OpenAI’s extended thinking, allowing the model to reason through harder problems before responding.

The model is designed for enterprise and developer use, not casual chatting. It targets teams building AI agents, automating knowledge work, and solving problems that require both reasoning and tool use. Google positions it as a workhorse, not a spectacle.

Market Context and Competition

Google’s delay coincided with broader market turbulence. The company watched Gemini 3.5 Pro get postponed, two celebrated researchers depart for OpenAI and Anthropic, and roughly $225 billion in market value evaporate in a single trading session. The pressure was real. So Google did what it does: iterative refinement and a measured rollout.

Gemini 3.5 Flash, the cheaper sibling released earlier, is already doing heavy lifting for developers who can’t wait. It offers most of Flash’s strength at lower cost. But Pro is for customers who need the best, and the wait until July signals Google is taking its frontier model seriously.

The model’s real test comes now—whether enterprise teams choose to build with it, and whether its reasoning and coding chops justify the wait.