Google's most powerful AI model was supposed to be in your hands last month. Alphabet CEO Sundar Pichai announced at Google's annual developer conference in May that Gemini 3.5 Pro was already being used inside Google and would roll out to customers the following month. That was June. The model still has not shipped.
Bloomberg, citing 10 current and former employees, reported on July 16 that the delay is specifically about coding. AI models are increasingly used to write, review, and debug software, and Gemini 3.5 Pro is not yet good enough at that job to meet Google's own targets. Google tried to address the gap in late June by updating the data used to train the model. The results fell short of what engineers hoped for.
The delay sent Alphabet's stock down roughly 4% in a single day. That is a meaningful signal from investors who have been patient with Google's AI spending for years and are now looking for delivered product, not promises.
The competitive timing is brutal. In the weeks that Google has been missing its deadline, OpenAI shipped GPT-5.6, Anthropic released Claude Fable 5, and Meta launched a new coding-focused model. On Google's own Android coding leaderboard, Anthropic's flagship took first place with an accuracy score of 84.5%, well ahead of Google's own models.
Sundar Pichai acknowledged as recently as May that Google was "a bit behind" on AI coding tools. The honest reason is structural. Unlike Anthropic or OpenAI, which build models that stand on their own, Google's Gemini must integrate with Search, Maps, YouTube, Google Workspace, and Cloud at the same time. Every release requires sign-off from multiple teams across a company of nearly 200,000 people. Former employees described the process as trying to quickly turn a massive ship.
There is a talent dimension too. In a single week in June, four senior researchers left Google for Anthropic and OpenAI, including one who worked directly on Google's AI coding efforts and another who worked on the foundational training stage that shapes what a model can do. The people who were supposed to fix the coding problem are now building for competitors.
Google's structural advantages have not disappeared. Its cloud business grew 63% in the first quarter of 2026. Its data resources and computing infrastructure are enormous. The lighter, faster version of its model, Gemini 3.5 Flash, is already live and is widely considered cost-effective for high-volume tasks. On multimodal work, handling text, images, audio, and video together, Google is still considered a strong option.
But coding is where the money is moving. Businesses are increasingly paying for AI tools that write software, review code, and automate technical work. That is where OpenAI and Anthropic have been winning deals. Google has been forming a dedicated coding team and reorganizing internally since April to close the gap. The reorganization started before the people leading it began leaving.
Alphabet reports second-quarter earnings on July 22. Investors will be watching Google Cloud revenue closely. If cloud growth slows at the same moment the flagship model is visibly late, the narrative around Google's AI position will harden in a way that will take more than a product launch to reverse.
For businesses currently choosing or re-evaluating their AI vendor, the practical takeaway is this: Gemini remains a solid choice for tasks that benefit from Google's network, particularly anything connected to Google Workspace or requiring large amounts of text processed cheaply. For software-writing tasks specifically, the options that have actually shipped are currently stronger. That could change when Gemini 3.5 Pro eventually arrives. But there is no new timeline, and Google has not announced one.