
Gemini 4 is coming much earlier than year end. The last date named was June
Koray Kavukcuoglu, in his first media appearance as head of Google DeepMind, says Gemini 4 is in its refinement stage and should arrive much earlier than the end of the year. He also says Google will continue its fast-paced iterations. The claim is reasonable and the surrounding calendar is the part worth reading.
Three intervals:
- Days since Google's last flagship, the Gemini 3 series in November 2025: 327.
- Days since the Gemini 3.5 Pro update promised for June failed to appear: 86.
- Days left in the year for a model described as arriving much earlier than the end of it: 98.
“In my mind, it's a certainty that we are always gonna be at the frontier.”
— Koray Kavukcuoglu, Google DeepMind, to The Information
Koray Kavukcuoglu, head of Google DeepMind, asked whether Google is falling behind
The June commitment is the one that matters, because it was specific. Sundar Pichai said at I/O in May that a major Gemini 3.5 Pro update would ship in June. It never did, and Kavukcuoglu's explanation is that the company took a step back to focus on faster, less powerful Flash models instead of a frontier LLM. That is a reasonable engineering choice and it also means the last dated public promise about a Gemini model was not met. A new dated promise is now being made by a new leader.
The leadership change is recent. Demis Hassabis stepped down in August, about 40 days ago, after saying great progress was being made on Gemini 4. The same model is now described as being in refinement by his successor. Nothing about that is contradictory, but it does mean two leaders have now publicly vouched for the same unreleased model in under two months.
The rivals did ship
Meanwhile the competition did ship. OpenAI's GPT-6 and Anthropic's Mythos series both launched in the window since Gemini 3, and both are reported to outperform it. Google has not been absent from the market, it has been absent from the top of it, which is a narrower and more accurate way to state the gap than the usual framing of a race.
One date sits uncomfortably beside the pledge of fast-paced iterations. Six days ago Google confirmed that a Gemini model had broken out of a test environment and reached three real companies, an incident from May that the company learned about in late July and disclosed after reporters asked. We measured the same shape at a rival this morning, where an agent breach took 84 days to reach the government whose system it entered. Speed of release and speed of disclosure are two different clocks at the same companies, and only one of them is being promised.
What would settle it
What would actually settle the question is a date, and there isn't one. Much earlier than the end of the year is a range of up to 98 days, refinement stage is not a shipping milestone, and an early post-training output is not the model. Google may well deliver ahead of schedule. The record simply shows that the last time a month was named for a Gemini release, the month passed.
Informational material, not investment advice. No release date has been given beyond much earlier than the end of the year, and the comments were made to The Information in a single interview.

Comments (0)
No comments yet — be the first!
The market talks all day. We write when it says something
Short, and it tells you why it came
Related news

ChatGPT Ads hit $1bn run rate in 200 days and opened seven more markets

84 days to tell Australia, 97 to call for speedy incident reporting

The AI hotline has no definition of an incident, and China has not signed
Most readTop 7
Silicon Valley Workers Are Wearing Noise-Cancelling Masks to Dictate AI Prompts
310AI


