Gemini 3.5 Pro release date: what happened to Google’s delayed model

Gemini 3.5 Pro update (checked 3 October 2026)

Google never released Gemini 3.5 Pro, and on 30 September 2026 it announced Gemini 4 Argon instead. Our Gemini 4 Argon guide covers its price, independent test results and availability.

The answer in brief

There is no Gemini 3.5 Pro release date, and there may never be one. Google promised the model for June 2026, missed that and a rumoured July date, then moved its effort to Gemini 4. It has not formally cancelled 3.5 Pro.

• What Google has said: its leaders have said it needed a larger Gemini 4 model to compete at the frontier, and chose to focus on its Flash models in the meantime.
• What Google’s pages show: the Gemini Pro page still carries a “3.5 Pro coming soon” label, but 3.5 Pro is not in the Gemini API model list.
• What you can use now: Google’s Flash models, the newest being Gemini 3.8 Flash.

For most marketers a delayed AI model sounds like a developer story. It isn’t. The Gemini family already decides how millions of people see brands in Google’s AI answers, so the state of its most capable model is worth understanding, even if you never open a developer console.

The sections below are kept as a dated record of how the delay unfolded in July, with notes where things have changed since.

When is Gemini 3.5 Pro coming out?

There is no confirmed date, and both reported ones passed without a launch. Leaks pointing to 17 July spread across dozens of outlets in early July, though nearly all of the coverage traced back to two publications, Geeky Gadgets and HackerNoon. The Bloomberg report published on 16 July, citing people familiar with the matter, described a model months behind schedule with no new date attached. The day passed without a launch.

Google’s next move made the position clearer. On 21 July it released three models and left Pro out: Gemini 3.6 Flash, pitched at token efficiency in coding and multimodal work; Gemini 3.5 Flash-Lite for high-volume tasks; and Gemini 3.5 Flash Cyber, a security-focused model restricted to governments and trusted partners. Google DeepMind product lead Logan Kilpatrick said the company is testing 3.5 Pro with partners and hopes to land it soon. Soon is doing a lot of work in that sentence.

The backstory explains how we got here. Google announced Gemini 3.5 at its I/O conference on 19 May 2026 and shipped the lighter Flash model the same day, while keeping Pro in a limited preview. From the stage, Sundar Pichai asked developers to “give us until next month to get it to you.” Next month meant June. June came and went, a Business Insider report moved the target to July, and July ended without it.

In July, Google DeepMind’s model page listed Gemini 3.6 Flash, 3.5 Flash-Lite, 3.1 Pro and 3.1 Deep Think as available, with 3.5 Pro marked coming soon. That label was still on the Gemini Pro page on 3 October 2026, but no model card, pricing page or launch post for 3.5 Pro has appeared, so every spec below remains reporting rather than fact.

Why did Google delay Gemini 3.5 Pro?

The official explanation is that Google wants more time with early testers and is using their feedback to refine the model before a wide release. It has also applied what it learned shipping the Flash models to Pro’s development.

Earlier reporting put a structural edge on it. According to HackerNoon, Google discarded the model’s original base, built on Gemini 2.5 Pro, and ran a fresh pre-training cycle on a Gemini 3 foundation instead, after internal testing showed weaknesses in multi-step maths and in keeping structure intact across long tool-calling chains. If accurate, that is not a polish pass. It is a rebuild, and rebuilds take weeks.

Bloomberg’s reporting points at coding specifically. According to people familiar with the matter, Google refreshed the model’s training data late in June to improve its coding skills, and the results disappointed. Ten current and former employees described frustration inside the company over losing ground to rivals. That echoes what Pichai admitted back in May, that Google was “a bit behind” the frontier on agentic coding. The scrapped-base account and the coding account are not mutually exclusive, but they come from different sources, and Google has confirmed neither.

Google’s own explanation came later. On Alphabet’s earnings call on 22 July, Sundar Pichai described 3.5 Pro as in testing, then told analysts that the next generation of frontier models needs much larger base models, and that Google would “need Gemini 4 as a larger base model” to compete there. In September, Google DeepMind’s Koray Kavukcuoglu said the company had taken a step back from 3.5 Pro to focus on its Flash models. Neither said 3.5 Pro was cancelled, although the analyst firm SemiAnalysis reported in August that it had been quietly dropped.

The timing is uncomfortable for Google either way. Alphabet’s share price came under pressure in late June, partly over concerns that the company is losing ground in AI and partly after it lost senior researchers to Anthropic and OpenAI. A flagship that slips past its own window, then a reported one, then a release day where three other models ship instead, does nothing to quiet that.

What you can use today: the Gemini Flash line-up

This is the part marketers should not skip. Pro may be missing, but the Flash models are live, and they are the ones doing the work that affects your brand right now. Gemini 3.5 Flash launched in May as the default in the consumer Gemini app and in AI Mode in Google Search, and it recently gained a built-in computer use tool that lets it carry out tasks in a browser-style environment. The 21 July release added Gemini 3.6 Flash, which Google says cuts token use by up to 17% while improving coding and multimodal performance, and Gemini 3.5 Flash-Lite, which has started appearing in AI Mode.

In plain terms, the AI answers a large share of your audience sees in Search are generated by the Flash tier, not by the unreleased Pro, and Google has spent July making that tier faster and cheaper rather than waiting for its flagship. If you have been holding off on Gemini until Pro arrives, you have been overlooking the models that are already shaping how customers find you.

That pattern held. Google shipped Gemini 3.7 Flash on 13 August and Gemini 3.8 Flash on 2 September; 3.8 Flash is its newest Flash model in the Gemini API and the newest model it has announced for AI Mode, for Google AI Pro and Ultra subscribers.

What Gemini 3.5 Pro was expected to bring

Google never published full specifications, so the following is what was reported before the model stalled. According to industry coverage, Pro was to have a two-million-token context window, double the one million Flash handles and among the largest in any production model to date. That is enough to hold thousands of pages, an entire content library or a year of campaign data, in a single request.

The headline addition is a “Deep Think” reasoning mode, which works through a problem in steps and checks itself before answering rather than responding on instinct. The mode has form: the existing Deep Think has scored 84.6 per cent on ARC-AGI-2 and took a gold-medal result at the 2025 International Mathematical Olympiad, though Google has not said how, or whether, it carries over to 3.5 Pro. Reporting indicated Deep Think would be reserved for Google’s top Ultra subscription, said at the time to cost around $250 a month. In the UK, Google AI Ultra now costs from £79.99 a month on Google’s UK plans page (checked 3 October 2026).

What the delay means for marketers

Set the release-date drama aside and one point matters for your strategy. The models that decide how brands appear in Google’s AI answers are live today and improving on their own schedule, and a more capable one will arrive when Google says so, not when the rumour mill does. Waiting changes nothing useful.

A few things follow from that.

First, optimise for AI answers now, not later. Answer engine optimisation, sometimes called generative engine optimisation, has stopped being a future concern. The Flash models already power AI Mode, so how your content is written, structured and sourced affects whether your brand gets cited in those answers today. Clear, direct responses to real questions, backed by genuine depth, are what these systems tend to surface.

Second, expect the bar to rise as more capable models arrive. Google built its next top model, Gemini 4 Argon, for long, multi-step work, and answer engines running models like it are likely to read far more source material before they answer, in the deeper research sessions where comparison and buying decisions get resolved. The more thoroughly an answer is researched, the more it rewards real expertise and comprehensive content over thin, keyword-led pages. Brands with authority on their subject stand to gain; those relying on shallow content have more to lose.

Third, measure your visibility in AI answers as its own metric. Appearances and citations inside AI Overviews and AI Mode rarely show up cleanly in standard click data, yet they shape perception and demand. Treating that visibility as a key performance indicator, alongside organic rankings, is the sensible response to a search experience that increasingly answers before it links.

None of this depends on which model Google ships next. It depends on acting while your competitors are still waiting.

What else launched while Google was waiting?

The rumoured date did not sit in a quiet month. GPT-5.6 Sol launched publicly on 9 July, and xAI opened Grok 4.5 to the public the same day. DeepSeek V4 reached general availability on 20 July, and its legacy API names were switched off four days later, forcing developers on the old models to migrate. Google’s own 21 July release added three Flash-tier models. Four frontier launches inside a fortnight is unusual even by 2026 standards, and Google is the only one of the pack whose flagship missed its window.

Where Gemini 3.5 Pro sits in the wider AI race

A slipped launch on a single model is minor in isolation. What makes it notable is the company it keeps. The frontier models from Google, OpenAI and Anthropic were bunching around the same stretch of summer, and each lab was watching the others’ dates as closely as its own. By October the comparison had moved on to OpenAI’s GPT-6 Astra, Anthropic’s Claude Opus 5.5 and Google’s own Gemini 4 Argon.

The lesson for anyone building a strategy on these tools is steadiness. No single lab is comfortably ahead, the lead changes with each release, and a few weeks of refinement often counts for more than being first. Plan around what a model does for your marketing, not around the headlines its launch generates.