Gemini 3.5 Pro: Release Date, Specs, and What's Confirmed

Google’s next flagship large language model is close. According to a wave of reporting and leaks circulating this week, Gemini 3.5 Pro is targeting a July 17, 2026 launch, after slipping from an earlier June window that Google set at its I/O developer conference in May. The most-repeated claims describe a 2 million-token context window and a new Deep Think reasoning mode — but almost none of it has been confirmed by Google in a model card, pricing page, or signed release post. This is a guide to what is actually known versus what is still rumor.

If you are new to the underlying technology, our explainer on what an LLM is covers the fundamentals these models build on.

What Google has actually said

Google announced the Gemini 3.5 Pro name at I/O on May 19, 2026, positioning it as the capability flagship above the already-shipped Gemini 3 family. At the time, the company signaled June for general availability. That target came and went. Google has since acknowledged the model remains in a limited preview for a handful of enterprise customers on Vertex AI, and pointed to a July launch window — while declining to attach a firm public date, publish benchmark numbers, or confirm specifications.

That is the extent of what is officially on the record. The far more detailed picture now circulating — the July 17 date, the context window, the pricing, the “architectural rebuild” — comes from third-party reporting and social-media leaks, not from Google. Treat every specific number below as unconfirmed until Google ships documentation.

What’s being reported

The leaks are unusually consistent, which is worth noting even though consistency is not confirmation. The recurring claims:

• A July 17 launch target. Multiple accounts describe an internal target of July 17, with the extra weeks since the June slip attributed to a new pretraining run rather than a simple polish pass.
• A 2 million-token context window. The headline capability in nearly every leak is a jump to a 2M-token window — double the 1M that has been standard on Google’s Pro tier — which would let the model hold very large codebases, document sets, or long agent histories in working memory. Our primer on what a context window is explains why that number matters.
• A “Deep Think” reasoning mode. Reports describe a dedicated Deep Think reasoning layer aimed at logic, mathematics, and multi-step problems — the same “think before answering” approach behind the broader class of reasoning models. Google already ships a Deep Think tier in the Gemini 3 line, so this would be an evolution rather than a first.
• Stronger frontend and design output. A cluster of leaks emphasizes better UI and frontend code generation, cleaner design “taste,” and stronger SVG rendering — a pointed response to the areas where rivals have been winning developer mindshare.
• More capable agentic workflows. Several accounts describe improved autonomous tool use and multi-step agent orchestration, aligning the Pro tier with the AI-agent push across the industry.
• Reported API pricing. Leaked figures put input pricing around $12–$15 per million tokens and output around $36–$45 per million — numbers that, if accurate, would sit well above the budget tiers but within range of frontier competitors. No official pricing exists.

Separately, reporting suggests Google is preparing adjacent models, including a “Nano Banana Pro” image generator and a Gemini 4 Flash speed tier — again, unconfirmed and not part of the 3.5 Pro launch itself.

Why the delay

The through-line in the reporting is quality control. Google set a June target at I/O, then pushed it after early enterprise testers flagged issues in reasoning and coding performance during the limited preview. Some accounts go further, describing a decision to scrap the prior base architecture in favor of a fresh pretraining run — which, if true, is the kind of change that explains a multi-week slip and a “rebuild” framing.

There is a strategic reading here too. Google is launching into a frontier that moved sharply in early July. OpenAI shipped its GPT-5.6 family — Sol, Terra, and Luna, and xAI released Grok 4.5, pitched as an “Opus-class” coding model at aggressive pricing. Shipping a Pro flagship that underperforms those on coding and reasoning benchmarks would be a costly unforced error. Waiting to get the numbers right — even at the cost of a public slip — is a defensible trade when the competitive set has just reset expectations.

Where it fits in the model race

Gemini’s distinctive advantage has never been a single benchmark; it is distribution. Google can drop a model straight into Search and a billion-plus everyday users — exactly what it did with Gemini 3.5 Flash, which now powers Google’s AI Mode in Search. A stronger Pro tier feeds the top of that funnel: the model enterprises license through Vertex AI and build products on.

The competitive framing is straightforward. On the capability axis, 3.5 Pro would be measured head-to-head against OpenAI’s GPT-5.6 Sol and Anthropic’s most capable Claude models, with coding and agentic performance the battlegrounds that matter most to paying developers. On the economics axis, the leaked pricing suggests Google is comfortable charging frontier rates for the Pro tier while keeping Flash cheap for scale — a barbell strategy that mirrors how the whole field now segments its lineups.

The 2M-token window, if it ships, is the most concrete differentiator on offer. Long context is where large document analysis, whole-repository coding, and durable agent memory live, and doubling the window is the kind of specification that shows up directly in what developers can build rather than in a leaderboard cell.

How to read the leaks

A word of caution is warranted, because the gap between leak and launch is where most predictions break. Launch dates slip — this model has already slipped once. Context windows and pricing quoted in preview rarely survive unchanged to general availability. And benchmark claims circulating before a model card exists are, by definition, unverifiable. The sensible posture is to treat July 17 as a plausible target, not a commitment, and to withhold judgment on capability until Google publishes numbers you can reproduce against your own workloads.

What it means

Gemini 3.5 Pro matters less as a single release than as Google’s answer to a frontier that shifted under it in early July. With GPT-5.6 and Grok 4.5 already shipping, Google is under real pressure to land a Pro flagship that is competitive on coding and reasoning — and its willingness to slip the date twice suggests it would rather be late than underwhelming.

Who’s watching most closely. Enterprise developers already building on Vertex AI have the most at stake: a 2M-token window and stronger agentic tool use would directly expand what they can ship, while a weak coding showing would push them toward rivals. Google’s Search and consumer surfaces are the quieter beneficiary — a better Pro tier eventually cascades down into the Flash models that serve everyday users.

What to watch next. Three things separate a real launch from more leaks: an official model card with reproducible benchmarks, a published pricing page on Vertex AI and the Gemini API, and confirmation of the context-window number in the docs rather than a screenshot. Until those land, Gemini 3.5 Pro remains the most-anticipated model of the summer — and one of the least confirmed.