At Google I/O in May, Sundar Pichai promised Gemini 3.5 Pro would ship in June. As of this week, that promise is about to be kept. Google’s flagship model in the Gemini 3.5 family is entering its final stages before general availability, and the capabilities on the table are worth paying attention to for anyone running serious data or AI workloads.

What’s Actually New

The Gemini 3.5 family splits into two tiers. Flash is already live and already good. Released at Google I/O on May 19, Gemini 3.5 Flash outperforms last year’s 1.5 Pro on most benchmarks at a fraction of the cost, running at $1.50 per million input tokens and $9.00 per million output tokens. For everyday tasks, Flash is the right tool.

Pro is where the story gets more interesting.

The headline capability is a 2 million token context window. That’s double Flash’s capacity and larger than any production frontier model currently available at scale. To put it in practical terms: 2 million tokens is enough to load an entire codebase, a year of customer support conversations, a full legal contract library, or tens of thousands of rows of financial data into a single AI session. The model then reasons across all of it at once, rather than requiring you to chunk documents and manage multiple calls.

The second headline feature is Deep Think reasoning. Google is positioning this as a step up in reliability for complex, multi-step problems. For data analysts working through ambiguous business questions, or any professional who needs the model to think through trade-offs rather than just generate an answer, this matters more than raw throughput.

Where Things Stand Right Now

Pro is not yet universally available. As of late May, Google opened limited access to Vertex enterprise customers, giving companies in regulated industries, financial services, and large-scale analytics an early look. Full general availability is expected any day in June.

This creates a practical decision point for businesses currently planning their AI stack. If your organisation is evaluating GPT-5.5 or Claude Opus 4 for data-intensive use cases, Gemini 3.5 Pro is now a serious third option, particularly where context length has been the bottleneck.

What This Means for Business

The jump from 1M to 2M token context sounds like a benchmark number, but it changes what is actually possible in practice.

For financial teams, you can load a year of quarterly reports, analyst calls, and industry news across dozens of companies into one session. Ask the model to surface correlations, flag risk signals, or draft a briefing. Previously this required careful document chunking and multiple API calls. At 2M tokens, it becomes one conversation.

For legal and compliance teams, you can process entire contract libraries in a single pass and surface clauses that contradict company policy or flag non-standard terms without a manual first pass by a lawyer.

For data teams, you can pull in schema documentation, query logs, business context, and data definitions simultaneously. The model can write SQL, explain results, and catch logic errors with full awareness of your environment, not just the snippet you pasted.

These are not hypothetical scenarios. They are exactly the kinds of tasks that enterprise customers on the Vertex limited preview are already running.

The Model Market Is Getting Competitive Fast

What is happening with Gemini 3.5 is part of a broader sprint. Microsoft launched seven in-house AI models at Build 2026 this week. OpenAI expanded Codex to non-developer roles. Every major platform is competing hard for enterprise AI spend, and the result is genuine improvement in what businesses can access at competitive price points.

For organisations, this is mostly a good problem to have. The challenge is no longer whether AI models are capable enough. The real challenge is building the internal skill and process discipline to take advantage of them.

That gap, between a capable model and a business that can actually use it well, is exactly what determines who gets ROI from AI and who does not.

What to Do Now

If your organisation runs on Google Cloud, enterprise Vertex customers can request early Pro access through their account team. For everyone else, the general availability announcement is expected within days.

If you are still working out how to evaluate AI models and decide which tools belong in your stack, that is a skills and strategy question as much as a technology one. The data literacy and AI courses at Enterprise DNA are designed to help data professionals and business teams make better decisions about AI, not just use the tools that are already familiar.

Gemini 3.5 Pro arriving this month matters. Context length has been a genuine constraint for enterprise AI work, and Google is about to significantly expand it.