You went to add gemini-3.5-pro to an API call this week because every headline pointed at July 17. It isn't there. Google let a third target for the Pro model slide past without a launch post or an API listing, and as of today it still hasn't shipped. Here's the part worth your time: almost everything you've read about it, the 2 million token context, the Deep Think mode, the price near fifteen dollars a million, even the July 17 date itself, is leak and reporting, not a Google announcement. What Google actually put its name to is one careful sentence about still testing. We went and pulled the confirmed apart from the circulating, because if you're planning a build around this thing, that line is the only part that pays off.
The short answer
Gemini 3.5 Pro was widely reported to land on July 17. It didn’t. That’s the third target it has slipped, and there’s still no model card and nothing in the public Gemini API. Everything you’ve read, the 2M context, Deep Think, the price near $15 a million, is leak, not launch. Here’s the line between what Google actually said and what’s just circulating.
The July 17 date was never Google’s
Let’s clear this up first, because it’s the whole story. Google did not announce a July 17 launch. It didn’t miss its own promise on that exact day, because it never made one. That date came from tech outlets and unnamed internal sources, and it spread because everyone wanted a peg to hang the coverage on.
So what’s true as of today? There’s no launch blog post. No model card. No pricing page. And crucially, if you open the Gemini API changelog, there’s no gemini-3.5-pro in it. The most recent public model is still gemini-3.5-flash, which shipped on May 19. Anyone who wrote code against gemini-3.5-pro this week was coding against a rumor.
That’s the honest headline. Not “Google launched a monster context window”, but “Google let a widely reported date pass and said almost nothing.”
The one thing Google will actually say
Here’s what Google has actually confirmed on the record, via statements picked up by Bloomberg and 9to5Google. One, it’s “currently testing 3.5 Pro, an upgraded Flash model, and other models with partners.” Two, the holdup is quality, particularly on coding. That’s it. No specs, no date, no price.
Back at I/O on May 19, the framing was softer: Flash was out, and Pro was “already being used internally”, with a rollout promised for “next month”. Next month meant June. June came and went. The team apparently reset the model somewhere in there, and the target drifted into July, then to the reported July 17, and now past it.
Why it slipped: coding
The reason is boring, which usually means it’s real. Google is behind on coding performance, and it decided a weak coding model wasn’t worth shipping into a market where the bar just moved.
Per the Bloomberg reporting, engineers updated the training data in late June to lift coding skill, and the results were disappointing. Reporting goes further and says Google scrapped the original base model and rebuilt it from the ground up, targeting the specific gaps early testers flagged. We can’t independently confirm the “scrapped and rebuilt” detail, so hold that one a little looser than the “still testing, coding is the problem” line, which Google itself has effectively owned.
The competitive picture is what makes a slip sting. Claude Fable 5 landed July 1. GPT-5.6 hit general availability July 9. Both are out, both are being benchmarked in public, and both are coding-focused. That leaves Google as the one major lab without a 2026 flagship in general production on the timeline its own executives floated.
What to build against right now
If Gemini 3.5 Pro was penciled into a roadmap, unpencil it. You can’t ship on a model that has no API, no price, and no date. Plan for “sometime”, not July.
If you need a Google model today, Gemini 3.5 Flash is real and live, and it’s genuinely fast. It just isn’t the frontier-tier Pro everyone’s waiting on. If you need frontier reasoning or hard coding help, the two models that actually exist this month are GPT-5.6 and Claude Fable 5, and you can hit both over their APIs right now. Honestly, I’d rather build against a model I can call than a spec sheet I can’t verify.
And when Pro does land? Re-check every number in this piece against Google’s own model card. The 2M context might be right. The price might be right. But “might be right from a leak” is a bad foundation for anything you’re going to bill a client for.
The honest read
I want Gemini 3.5 Pro to be good. A strong third frontier model keeps the other two honest on price, and Google’s long-context work has been legitimately ahead. So this isn’t a dunk.
It’s a correction. The internet has been treating a leaked date and a leaked spec sheet as a launch, and it wasn’t one. What shipped is Flash, in May. What’s confirmed about Pro is that Google is still working on the coding, and that’s the entire on-record story. Everything with a number attached is somebody’s report, not Google’s word. Wait for the model card. Then test it on the prompts that actually matter to your work, and let the leaderboard tweets sort themselves out.
Sources: Bloomberg, 9to5Google, Google (I/O post, May 19), Gemini API changelog and TechTimes, July 2026. Google confirms it is still testing 3.5 Pro and cites coding quality as the delay; the July 17 target date, the 2M token context window, the Deep Think mode and all pricing figures are as circulating via reporting and leaks, are not from an official Google model card, and are not independently verified. The May 19 Flash launch and the “next month” line are from Google’s own I/O post.
Frequently asked questions
Is Gemini 3.5 Pro out yet?
No. As of July 18, 2026, Google has not launched Gemini 3.5 Pro. There is no model card, no pricing page, and no gemini-3.5-pro entry in the public Gemini API. The widely reported July 17 target came from third-party reporting, and it passed with nothing shipped. The only Gemini 3.5 model actually released is Flash, back on May 19.
What is Gemini 3.5 Pro's context window?
The number doing the rounds is 2 million tokens, double the 1M on Gemini 3.5 Flash. Treat it as a leak. Google has not published a model card for Pro, so the context window, like the rest of the spec sheet, is reporting rather than a confirmed figure.
How much will Gemini 3.5 Pro cost?
Unknown officially. Estimates in circulation land near $15 per million input tokens and somewhere from $36 to $60 per million output, with the Deep Think reasoning mode reportedly gated behind a pricier consumer tier. None of that is confirmed by Google, so budget nothing around it yet.
Why is Gemini 3.5 Pro delayed?
Google says it is still improving the model, and coding is the sticking point. Per Bloomberg, an attempt in late June to retrain on better coding data gave disappointing results, and reporting says Google scrapped the original base model and rebuilt it from scratch. The official line is that the company is testing 3.5 Pro with partners while shipping other models in the meantime.
What can I use instead while I wait?
The two frontier models that actually shipped this month are GPT-5.6, which reached general availability on July 9, and Claude Fable 5, out since July 1. Gemini 3.5 Flash is also live now if you specifically want a Google model today. Any of those is a real option. Gemini 3.5 Pro is not one, until Google posts it.
