Google's most powerful model of the year is finished, demonstrated, and priced. There is just one problem: almost no one can use it yet. As of June 21, Gemini 3.5 Pro remains locked behind a limited preview open only to a handful of select Vertex AI enterprise customers, even as the calendar marches toward the general-availability window CEO Sundar Pichai promised the world from the Google I/O stage a month ago.
The gap between announcement and availability has become the most-watched story in frontier AI this month, and it tightens by the day.
What Pichai Promised
When Pichai introduced Gemini 3.5 Pro at Google I/O on May 19, he framed it as imminent rather than aspirational. "We're also hard at work on 3.5 Pro," he told the audience. "It's already being used internally, and we look forward to rolling it out next month." When pressed on timing, his on-stage answer was blunter still: "Give us until next month to get it to you" — a line that, by multiple accounts, drew audible groans from a crowd that had hoped to leave Mountain View with access in hand.
That "next month" is now. June is Google's self-imposed deadline, and the company has not budged from it publicly. But it has not shipped, either. Gemini 3.5 Pro has not reached the consumer Gemini app, Google AI Studio, or the general developer API. The only people touching it outside Google are enterprise customers who have been allowlisted into a Vertex AI preview — and even they are operating under preview terms, not production guarantees.
The Specs That Matter
The reason the wait is generating real heat is that the confirmed specifications put Gemini 3.5 Pro at the frontier of what a production model can do.
The headline number is the context window: 2 million tokens. That is double the 1-million-token window of Gemini 3.5 Flash, which launched in May, and it stands as the largest input context of any production frontier model. In practical terms, a 2M-token window can hold the equivalent of multiple full-length books, an entire codebase, or hours of transcribed audio in a single prompt without truncation — the kind of capacity that reshapes how enterprises think about retrieval, summarization, and long-horizon agentic work.
Alongside the window sits "Deep Think," a reasoning mode that lets the model spend additional compute working through hard problems — extended chain-of-thought aimed at advanced math, science, and coding — rather than answering quickly. It is Google's clearest statement yet that it intends to compete on deliberate reasoning, not just speed and scale. The model also carries frontier multimodal understanding, handling text, images, and other formats natively.
The Pricing Question
What Google has not confirmed is the price, and that has become its own subplot. Leaked figures circulating among developers and enterprise buyers put Gemini 3.5 Pro at roughly $15 per million input tokens and $60 per million output tokens — a figure that should be treated as unconfirmed until Google publishes an official rate card.
If accurate, that pricing would land at roughly ten times the cost of Gemini 3.5 Flash, mirroring the Pro-to-Flash ratio of previous generations. For teams that scaled their economics around Flash, a 10x jump is not a rounding error; it is a budgeting decision. The reporting also suggests the extended Deep Think reasoning mode will be gated to Google's $250-per-month AI Ultra tier, which includes early access to the model, while standard AI Pro subscribers at $20 per month get the 2M context window and full model capability but not the extended reasoning mode.
That tiering tells you how Google is thinking about the model: the raw capability is broadly available, but the most compute-hungry reasoning is reserved for the customers willing to pay top-tier subscription rates.
Why This Matters
The delay is not happening in a vacuum. Gemini 3.5 Pro is Google's answer to a market that has not slowed down to wait for it. OpenAI's GPT-5.6 and Anthropic's latest Claude models are already in the field, and every week that Pro stays in limited preview is a week competitors get to define the frontier without Google's strongest entry on the board.
That dynamic is exactly why the June general-availability window carries so much weight. Pichai committed to it personally, on stage, at the company's biggest event of the year. Missing it would not merely be a slipped ship date — it would be a credibility hit at a moment when Google is trying to convince enterprises that it can execute at the same cadence as its rivals. The company has spent the past year arguing that Gemini's reach — Pichai cited 900 million users in his I/O keynote — makes it the platform to build on. A blown deadline undercuts that argument.
There is also a quieter signal in how Google is staging the rollout. By seeding Vertex AI enterprise customers first, Google is prioritizing the buyers who generate revenue and reference deals over the consumer base that generates headlines. It is a defensible sequencing choice, but it leaves the public-facing Gemini app — the surface most people actually associate with the brand — waiting in line.
What to Watch
The clock is the story now. Google has roughly nine days left in June to honor the general-availability commitment Pichai made in May. Prediction markets tracking a "released by June 30" outcome have hovered around even odds, leaning slightly yes — a striking level of uncertainty for a model the company has already demonstrated and priced internally.
Watch for three things before the month closes: an official pricing announcement that either confirms or resets the leaked $15/$60 figures; expansion of access beyond Vertex AI into AI Studio and the consumer Gemini app; and any language from Google that softens "June" into "this summer." The first two would signal that the deadline holds. The third would signal that the frontier's most anticipated launch has slipped — and that the groans from the I/O audience were prophetic.
“Give us until next month to get it to you.”— Sundar Pichai, CEO, Google and Alphabet