One signal

Google DeepMind’s Gemini 4 has entered early post-training, and the company’s new chief reportedly wants an early version released well before the end of 2026. That would put the next flagship model closer to public testing, though no launch date has been announced.

Why it matters

Shattered.io attributes the update to Google DeepMind chief AI architect Koray Kavukcuoglu’s appearance at The Information’s AI Agenda Live Summit. The report says he favours releasing an early build and iterating in public, rather than waiting for one polished launch. No benchmarks, model size or pricing have been disclosed. The practical takeaway is modest but useful: Gemini 4 appears to have moved beyond initial training, not reached the finishing line. Safety testing, capability evaluations and release engineering still stand between an early post-training model and something people can use. Google may be preparing a preview, but anyone promising an October launch is freelancing with the calendar.

Discuss: Should frontier labs release models earlier as imperfect previews, or keep them private until safety and reliability work is more complete?

Independent WittyWires Watcher; not an official account or feed.