Google Promised Us Its Best AI Model in June. It’s Still Not Here

Google’s had one of those months where you genuinely can’t tell if things are going great or falling apart, depending on which part of the story you’re looking at.

On one side, Gemini keeps sneaking into more places than anyone really expected — your browser, your desktop, apparently your fridge now too. On the other side, the one model everybody was actually waiting for just… hasn’t shown up. Not late by a week. Late by months, with no new date attached.

Both of these are happening at the same time, which is a little funny if you sit with it for a second.

So where’s Gemini 3.5 Pro, exactly?

Back in May, on stage at Google I/O, Sundar Pichai told the room that Gemini 3.5 Pro — Google’s flagship, top-of-the-line model — would be out in June. That was the promise, in front of a room full of developers who were taking notes.

June came. Went. Nothing.

Bloomberg ran a piece on this last week that’s worth actually reading, not just skimming the headline of. Ten current and former Google employees told them the holdup is mostly about coding — the model just wasn’t good enough at it internally, especially with OpenAI and Anthropic breathing down Google’s neck on that exact front. At some point in late June, Google apparently went back and reworked the training data specifically to fix the coding weakness. And by more than one account, it still wasn’t enough.

What makes this more than a “whoops, we’re a bit late” story is the mess underneath it. According to the reporting, Google’s had several teams — DeepMind, Cloud, Android, Search — all separately building overlapping coding tools, kind of stepping on each other without meaning to. There’s apparently a real push happening right now to pull all of that together under something called Google Antigravity, with one of the company’s chief AI architects personally getting involved to untangle it.

And here’s the detail that actually stopped me for a second: about 75% of new code shipping at Google right now is AI-written, up from roughly half just several months back. That’s Google leaning hard on AI internally, at scale — and yet, per the same report, engineers are running into compute capacity limits when they try to actually use the tools. Kind of an odd spot to be in: pushing AI coding company-wide while your own best coding model isn’t ready to leave the building.

Alphabet’s stock dipped a few percent the day the story dropped, so investors clocked it too. Google’s line, officially, is that 3.5 Pro is being tested with select partners right now, alongside an upgraded Flash model. No public date. Nothing promised this time.

While that’s stuck, everything else is moving fast

Here’s what makes the timing feel almost backwards. While the big flagship model sits in testing limbo, the rest of the Gemini lineup has been spreading out at a pretty aggressive pace.

Chrome now has an “Ask Gemini” button — small sparkle icon, top corner — rolling out to UK users on Mac and PC. It can read across several open tabs at once, which is honestly a nice trick if you’re comparing options or piecing together research from a dozen tabs you should’ve closed an hour ago. It also reaches into Gmail, Calendar, YouTube, and Maps — so you can jump to a specific moment in a video or draft an email without ever leaving the page you’re on. Riding along in that same update: Nano Banana, a text-to-image tool sitting right in the browser sidebar.

On the desktop side, Google shipped Gemini Spark for macOS, which goes a step past chatting — it’s meant to actually act on your local files and desktop tasks. Spark’s also picked up the ability to quietly track things in the background: a sports score once the match wraps up, a stock hitting a price you set, a story that keeps developing over a few days. Less “ask and get an answer,” more “keep watch and tell me when it moves.”

And then, because 2026 apparently demanded it — Samsung announced Gemini is coming to one of its new fridge lines. Make of that what you will. It says something about how far Google’s willing to stretch this brand.

Wait, which Gemini are we even talking about?

Honestly, this part trips people up constantly. Right now, Gemini 3.1 Pro is generally seen as the strongest reasoning model Google has out. Gemini 3.5 Flash, released back in May, is the newer and cheaper option — built around roughly a million-token context window, which makes it a solid fit for long documents, contract review, or research jobs where chunking text manually is a pain. Gemini Omni, also unveiled at I/O, is the multimodal generation play, starting with video.

The flagship — 3.5 Pro — is still the missing piece. So if someone says “the latest Gemini,” they could genuinely mean three or four different things depending on the day, and that’s less a Google screwup and more just a symptom of how fast this whole field is moving right now.

Does any of this actually matter for people using these tools day to day?

If you’re running a small business or working solo and leaning on Gemini already, none of this is really cause for alarm. The stuff people actually use daily — Flash for quick tasks, the Workspace integration, the Chrome assistant — is shipping fine and getting better. The delay is specifically about the top-tier reasoning and coding model, which matters a lot if you’re doing serious technical work, and a lot less if you’re mostly drafting emails or summarizing calls.

The bigger story here is competitive, if you ask me. Google publicly admitting its flagship isn’t good enough yet at coding — and having that frustration leak out from inside the company — while Anthropic and OpenAI keep winning that exact fight, tells you nobody’s holding a comfortable lead for very long right now.

Bottom line

Gemini’s having two completely different months at the same time. As something woven into browsers, desktops, and now apparently kitchen appliances, it’s spreading faster than almost anything else out there. As a frontier model trying to keep up with the best coding systems on the market, it just missed its own deadline in public.

Both things are true. Neither cancels the other out. If you want an honest snapshot of where this whole race actually stands in mid-2026, this is probably it — nobody’s comfortably ahead, and everyone’s shipping in public, mistakes and all.