Google's new Gemini 3.5 Flash, explained — and the one that's not out yet

At its big yearly event (Google I/O) on 19 May 2026, Google announced a new AI model: Gemini 3.5 Flash. If you use the Gemini app, or see AI answers at the top of your Google Search results, this is already affecting you — gently and, probably, positively. Here's what actually happened, in plain English, and what you need to know about the part that isn't ready yet.

Gemini 3.5 Flash is Google's new everyday AI model — and from 19 May 2026, it became the model powering the answers you see in the Gemini app, in Google's AI search results (called 'AI Mode'), and in the developer API. You probably didn't have to do anything. It was just switched on, everywhere, because Google controls all of those surfaces.

What's new about it? According to Google, Gemini 3.5 Flash is faster and more capable than anything Google offered before at this price tier — and Google goes further, claiming it actually beats the previous bigger model (Gemini 3.1 Pro) at certain tasks, like coding and following complex instructions. The headline number Google cited is 76.2% on something called Terminal-Bench 2.1 — a test that measures how well an AI handles programming tasks in a realistic setting. That's Google's own figure, so we'd call it encouraging rather than confirmed; independent researchers will run their own tests in the coming weeks.

Google also talked at I/O about Gemini 3.5 Pro — a bigger, more powerful model aimed at harder tasks. The catch: it wasn't actually released. In the announcement, Google said Pro was 'already being used internally' and that the team 'look forward to rolling it out next month.' So as of launch it isn't available to you or to people building apps with it — it's an internal model with a 'next month' target. If you've seen 'Gemini 3.5 Pro' mentioned and can't find it, that's why: it was announced, not launched.

Here's the honest bit, and it's worth knowing before you get excited: Google announced Pro with almost no details. It didn't say how big its memory (context window) is, didn't share any benchmark scores, and didn't give a price. All we really have is the name and the 'next month' promise. So treat 3.5 Pro as 'coming soon, details to follow' rather than a model you can plan around today.

Here's a simple comparison to help you see the difference:

For most people using Gemini or Google Search: Flash is what you've already got, and it's better than before. Pro is something to keep an eye on — Google just hasn't shared the details yet.

If you just use Gemini or Google Search, the change is probably a quiet improvement you'll notice as slightly sharper or faster answers. You don't need to opt in or update anything — it's already running. If you build apps using Google's AI API, there are two things to check. First, make sure you're getting the capability improvements you were hoping for (Google's benchmark claims are promising, but independent verification is still coming). Second — and this is important — check your pricing. Official pricing for Flash is $1.50 per million input tokens and $9 per million output tokens (with cached input cheaper, at $0.15). Google describes Flash as costing 'less than half' of other top-tier models — but that's a comparison to the big flagship models, not a promise that it's cheap in absolute terms. For an app that produces a lot of text, $9 per million output tokens adds up, so factor it into your budget.

One detail that's easy to miss: Google didn't just upgrade a model — it upgraded the model that powers a huge portion of everyday AI interactions. The Gemini app has tens of millions of users. Google Search's AI Mode reaches hundreds of millions. When a new model becomes the default at that scale, tiny improvements in accuracy or speed multiply into a huge difference in aggregate. That's part of why Google's decision to ship Flash first — rather than waiting for Pro — makes sense as a product move: get the better model into the hands of the most people, as fast as possible, and let Pro be the headline for when it's actually ready.

Two things to keep an eye on. First, independent benchmark results for Flash — when third-party researchers publish their own tests (usually within a few weeks of a launch), you'll get a clearer picture of whether Google's claims hold up outside Google's own testing. Second, when Gemini 3.5 Pro actually arrives — 'next month' from 19 May has so far not materialised for most users; when it does, that will be a genuinely significant upgrade for people doing complex work. Until then, Flash is the real upgrade on the table, and it's already running wherever you use Google's AI.

It's also worth keeping a casual eye on what AI Mode in Search actually feels like over the next few weeks. That's the biggest real-world test of any AI model: not a lab benchmark, but millions of ordinary questions getting real answers in real time. If responses feel sharper, more accurate, or faster, that's Flash working. If you notice something going wrong — wrong facts, weird refusals, odd tone — that's useful feedback too. The best test of any AI model is simply using it, and most of us already are.

A final note on the Pro side of the story: when Gemini 3.5 Pro does arrive, Google is positioning it as the heavier, more powerful option for genuinely complex tasks — writing long documents, analysing big datasets, coding large projects. The honest caveat is that Google hasn't yet said how much more capable it is: no benchmark scores, no memory (context window) size, no price. So treat it as 'the power-user model to watch' rather than something with known specs. Flash is the everyday upgrade you already have; Pro is the promised next step, and we'll know what it's really worth when Google actually ships it and shares the numbers.