Google's Flagship Model Gemini 4 Enters Post-Training Phase; DeepMind Head Hopes to Release It as Early as Possible This Year
nashnova research
Google DeepMind head Koray Kavukcuoglu confirmed that flagship model Gemini 4 has entered post-training — the fine-tuning stage for reliability — and is targeting release well before year-end. This means Google is racing to close a gap it has now publicly admitted.
How far along is Gemini 4?
Gemini 4 has finished foundational training and moved into post-training — the phase where a model is fine-tuned for reliability and safety. In plain terms = training made it smart; this stage makes it controllable.
The model is already in use among Google's internal engineers, powering its AI coding tool Antigravity. Safety testing and guardrail development are running in parallel.
This means → Gemini 4's core capabilities are largely set. The remaining work is calibration and safety clearance, not building from scratch.
Why was Gemini 3.5 Pro skipped?
CEO Sundar Pichai promised a Gemini 3.5 Pro release in June at the May I/O conference. That version never shipped.
Kavukcuoglu explained: after Gemini 3 and 3.1, the team "took a step back" to push smaller, cheaper Flash-series models. "The priority was maximizing learning velocity."
In plain terms = Google decided that forcing out a half-step upgrade was less valuable than running smaller models to gather data and folding those lessons straight into the next flagship.
Where are the rivals now?
Anthropic launched its Mythos series this spring and completed an upgrade this month. OpenAI began rolling out its GPT-6 series in early September.
Kavukcuoglu openly acknowledged that both competitors' current state-of-the-art models outperform Google's existing products.
This means → Gemini 4 is not a routine iteration. It carries the burden of proving Google can return to the top tier. Ship date and benchmark performance will directly shape the market's verdict on Google's AI standing.
What signal is the new leader sending?
This was Kavukcuoglu's first public appearance since taking the top role at Google DeepMind, delivered at The Information's AI summit.
He expressed strong confidence in his team: "I one hundred percent believe in the team. In my view, it is inevitable that we remain at the frontier."
This reflects a post-leadership-transition urgency: Google needs the outside world to hear one clear message — falling behind is temporary; the catch-up is already underway.
市场有风险，内容仅供研究参考，不构成投资建议。