The 17 July date, widely reported as the launch of Gemini 3.5 Pro, passed without any public release. The model remains confined to a limited preview for a few enterprise customers, and an August window is now being floated. Three missed deadlines since May. A game-changer: Google has confirmed that a custom Gemini model will power Apple's future Siri.
Last Friday, we told you about the delays piling up for Gemini 3.5 Pro, with a target set for 17 July. That date has passed. Google's flagship model is still not publicly available, and the matter takes on a new dimension with Apple entering the equation. A status check.
Counting the missed deadlines
Let's recap the timeline, because it has become telling. On 19 May 2026, Sundar Pichai announced on the Google I/O stage that Gemini 3.5 Pro would arrive "next month". June came and went with no release. Google then cited July, invoking quality adjustments. A 17 July target was widely reported, to the point that prediction markets gave it roughly a 62% probability. 17 July passed, with no launch.
Today, the model remains in a limited preview on Google's enterprise platform, accessible to a small number of approved customers. No official spec sheet, no pricing page, no public model identifier in the documentation. An August window is now being floated, with no more guarantee than the previous ones.
Nearly all the specs circulating about Gemini 3.5 Pro come from leaks and anonymous sources, not official announcements. The 2-million-token context window, the deep reasoning mode, the rumoured pricing around $15 input and $60 output per million tokens: all of this remains reported, never confirmed by Google. And the leaks contradict each other. In mid-July, one source claimed the model beat Fable 5 and GPT-5.6 in internal evaluations, while other tester accounts described it as lagging on code and long context. Impossible to settle until an independent third party gets access.
The Apple factor changes the stakes
Here's the element that turns a simple industrial delay into a heavier matter. Google has confirmed that a custom-built Gemini model will power Apple's rebuilt Siri. In other words, the voice assistant on hundreds of millions of iPhones will partly rely on the technology Google is currently struggling to finalise.
This is a huge commercial win for Google, placing its technology at the heart of its main consumer rival's ecosystem. But it's also considerable pressure. An internal delay can be forgiven; a delay that affects Apple's schedule costs far more, in reputation and in the contractual relationship alike. Google is no longer just disappointing impatient developers: it is now keeping a partner waiting that has built part of its roadmap on this technology.
What Google has delivered nonetheless
It would be unfair to sum up the situation as a failure. Gemini 3.5 Flash, the fast, cost-effective version of the family, has been available since May and does a remarkable job. It outperforms Google's previous top-tier model on code and agent tasks, while producing tokens roughly four times faster, at around $1.50 input and $9 output per million tokens.
This observation also sheds light on the strategy: Google shipped its fast model and held back its flagship, the exact opposite of the usual logic of leading with your best product. For a developer, the pragmatic advice remains the same: build today on what exists and works, and plan for a switch if the high-end finally delivers on its promises. Building an entire architecture around an unconfirmed date would be unwise.
What to take away
Three missed deadlines in two months do not mean Google has lost the race. The company has world-class research, colossal resources, and an already very solid mid-range model. But repetition eventually produces an effect of its own: each missed date erodes a little more of the credibility of subsequent announcements, turning anticipation into scepticism.
The real question is no longer really when Gemini 3.5 Pro will ship, but what it will have to prove upon release. After such a delay, a merely good model would be seen as a disappointment. Google has put itself in a position where only a genuinely superior model will justify the wait. It's a risky bet, and it will likely play out in August.