What Google actually delayed
In late June, Google changed the data used to train Gemini in a push to lift its coding scores. The results came back disappointing, and the flagship Gemini 3.5 Pro that many expected at the May developer conference stayed in limited enterprise preview. This week The Information and Bloomberg reported the slip runs to months, not weeks.
Google did not deny it. The company said it is testing Gemini 3.5 Pro, an upgraded Flash model and other systems with partners, and is discussing the model's capabilities and testing standards with the US government. What it did not give was a date, and the absence of one is the part that matters.
Why a coding miss hurts more than a general one
A model can be excellent at reasoning, writing and analysis and still miss on code, because coding rewards a narrow, verifiable kind of correctness that does not average out. Google's shortfall sits precisely there, on coding and long, multi-step tasks, and DeepMind staff have reportedly flagged internally that the company lacks a credible commercial coding product for the businesses building AI development tools.
That is the capability the market is buying right now. The fastest-growing AI spend is in coding assistants, agents and automation, and those buyers pick the model that writes and fixes code most reliably. Missing on coding is not missing on a feature; it is missing on the use case that currently pays.
The number the market put on it
Alphabet shares fell about 4.4 percent on the reports, erasing in the region of $225 billion of market value in a single session. That is not a verdict on Google's research depth, which remains among the best funded on earth; it is the market repricing a timeline it had taken for granted.
The competitive frame is what gives the number teeth. OpenAI's GPT-5.6 and xAI's Grok 4.5 are already in the market as coding-forward models, so every month Gemini 3.5 Pro stays in preview is a month rivals convert developer habits into defaults. Delays compound when the thing delayed is the thing customers are choosing on.
What this changes for a European buyer
If your plan assumed a specific Gemini coding capability by a specific quarter, that assumption now has no date behind it. For a European team weighing Google, OpenAI, Mistral or an open-weight model for an internal developer platform, the safe default just became the one with the visible gap on your core task.
It also argues for not marrying a single vendor before the capability ships. Building on a model you can test today, while keeping the integration thin enough to swap, costs a little now and saves a rebuild later. In euros, the expensive mistake is committing engineering quarters to a roadmap that a lab can move without telling you.
The discipline the delay rewards
Treat 'coming soon' from any lab as a capability you do not have yet, and design around what you can run this week. That is not cynicism about Google; it is the same rule that protects you when OpenAI or xAI slips next.
Google can still ship a strong model and close the gap, and given its resources it probably will. The owners who come out ahead are the ones who did not restructure a product line around a launch date that was never theirs to promise.
Read next: The Best Voice AI Is Not the One You Can Deploy | Google Scrapped a Near-Finished AI Model
