Brin takes over Gemini as 3.5 Pro is reportedly shelved

Share
Brin takes over Gemini as 3.5 Pro is reportedly shelved

Google's most delayed flagship model may never ship — and the co-founder is back in the room to fix what went wrong.

Sergey Brin has taken direct oversight of Google's Gemini team as reports say the long-delayed Gemini 3.5 Pro has been shelved. Chinese outlet QbitAI reported the co-founder stepped in urgently, echoing the Financial Times' August 6 account of a reorg that cements Brin's returning influence — he now sits near new DeepMind operations chief Koray Kavukcuoglu, and control of Google's AI effort is shifting from London back to Silicon Valley. The cancellation itself is unconfirmed: a SemiAnalysis report, relayed by IT之家, says Google may have internally shelved 3.5 Pro to clear the runway for a Gemini 4 push, but Google has not commented and still lists the model as "coming soon" on its Gemini page.

The context makes the rumor credible. A Fortune investigation published Monday found Gemini 3.5 Pro — unveiled at I/O in May with a promise to ship "next month" — missed its June and mid-July deadlines and remains unreleased, with engineers blaming a decision last summer to deprioritize coding while Anthropic and OpenAI raced ahead on agents. Staff described 60-hour weeks and burnout, and the leadership vacuum was literal: Brin was visible on the floor, Hassabis was not. The talent flight has been brutal — Gemini co-lead Noam Shazeer left for OpenAI in June, AlphaFold co-inventor John Jumper went to Anthropic, and Jeff Dean is departing after 27 years to found Discovery Loop with three other senior researchers. Demis Hassabis has moved up to chairman and Alphabet chief scientist, and SemiAnalysis now ranks Gemini eighth on the Artificial Analysis Intelligence Index, behind open-source Chinese labs. Brin taking personal command is the strongest signal yet that Google treats this as a code-red — but a founder in the trenches doesn't fix a shipping problem by itself.


The UK is betting its £500 million Sovereign AI programme on a 30-person London startup. The Financial Times this week profiled Cosine, the lab selected to train "Lumen Sovereign" — billed as Britain's answer to OpenAI — entirely on Isambard-AI, one of Europe's most powerful supercomputers, so that UK organizations get frontier capability without foreign dependence. The FT's reporting questions whether a team whose 2024 seed round was smaller than one San Francisco hire can credibly build a frontier rival, though Cosine points to its post-trained coding agents outperforming GPT-5 and Claude on software-engineering benchmarks, and Vodafone Business joined the coalition in June. Small labs have punched above their weight before — DeepMind and Mistral both started tiny — and government compute gives Cosine a moat money alone can't buy. Whether 30 people can hold it is the open question.

What to watch: whether Google confirms the 3.5 Pro cancellation and accelerates Gemini 4 — and whether Brin's return changes the lab's shipping record by year's end.

Brin is back in the trenches at 52 — does founder mode actually fix a lab that missed three deadlines? Tell us in the comments.

Sources: QbitAI · Fortune · The New York Times · IT之家 (via Sina) · Financial Times · Cosine (GitHub)