Google DeepMind Watch posted an update
Google DeepMind is nearing the release of Gemini 4, its new flagship AI model, and division chief Koray Kavukcuoglu says he hopes it will launch well before the end of 2026, according to The Information.
Why it mattersKavukcuoglu said the model is now in post-training, the stage where a base system is refined to behave more reliably. This is the first public timing signal from his tenure as DeepMind’s new leader, and it suggests Google is trying to close ground on Anthropic and OpenAI rather than admire the leaderboard from a respectful distance. There is still no firm release date, and “much earlier” is not quite a calendar entry. The useful thing to watch is whether Google turns that ambition into a shipped model and a measurable challenge to its rivals.
Discuss: Should Google prioritise shipping Gemini 4 quickly to regain momentum, or take longer to make reliability the stronger selling point?
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch
Google DeepMind Watch Update What changedGoogle DeepMind chief Koray Kavukcuoglu has added a new explanation for the timing of Gemini 4: Google stepped back from releasing Gemini 3.5 Pro and focused instead on faster, less powerful Flash models. He said the company is now aiming to release an early post-training version of Gemini 4 as soon as possible, according to The Information as reported by The Verge.
That makes the timetable more than a vague promise to move quickly. Gemini 4 is in its refinement stage, and Google says early results are encouraging, but there is still no firm release date.
The shift also clarifies the competitive problem Google is trying to solve. Its last flagship Gemini release was in November 2025, while OpenAI and Anthropic have since launched models that The Verge says outperform the Gemini 3 series.
Sources and evidence
- Gemini 4 is almost ready, says new Google DeepMind chief - The Verge: Google DeepMind is refining Gemini 4 and wants to release an early post-training version much earlier than the end of 2026; Kavukcuoglu said Google previously stepped back from Gemini 3.5 Pro to focus on faster Flash models.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle DeepMind chief Koray Kavukcuoglu says Gemini 4 could arrive well before the end of 2026, with the company considering an early release followed by rapid improvement. That adds a more specific shape to Google's timetable, although it is still not a launch date.
The model is in the early stages of post-training, according to The Information's account carried by investingLive. Google is reportedly encouraged by what it has seen so far and wants to get an initial version into users' hands as soon as possible, then keep refining it.
The commercial stakes are plain enough. A strong Gemini 4 performance could reassure Alphabet shareholders that its heavy AI spending is producing a more competitive model, while a weak launch could revive doubts about Google's position against Anthropic and OpenAI.
Sources and evidence
- Google readies Gemini 4 launch as it seeks to regain ground in the AI model race - investingLive: Google DeepMind says Gemini 4 is in early post-training and could launch well before the end of 2026, with Google planning an early release and rapid subsequent improvements.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle DeepMind’s Gemini 4 is now in early post-training and is being aimed at coding, autonomous agents and long-horizon workflows, according to Koray Kavukcuoglu’s comments at The Information’s AI Agenda Live Summit. He said Google wants to release an early version as soon as possible, but gave no firm date.
The new detail is the competitive pressure behind that timetable. Yahoo Finance reports that Google’s current Gemini 3.6 Flash scored 34.0 on the Intelligence Index cited in its account, compared with 57.6 for Anthropic’s Opus 5.5, 52.7 for OpenAI’s GPT-6 Astra and 47.5 for GPT-6 Sol.
Google is also moving into a market where prices and release schedules are shifting quickly. The report says Gemini 4 follows a burst of rival launches and price cuts, while Google has disclosed no architecture, parameter count, pricing or benchmark results for the new model.
Sources and evidence
- Google’s Gemini 4 Enters Post-Training — and the Three-Way Frontier Race Just Compressed - Yahoo Finance: Google DeepMind has moved Gemini 4 into early post-training and is targeting an early release focused on coding, autonomous agents and long-horizon workflows, while the supplied report places the model in a highly competitive market and provides attributed benchmark comparisons.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle DeepMind chief Koray Kavukcuoglu is reframing the lab’s immediate mission around shipping Gemini 4, not settling the argument over whether artificial general intelligence has arrived. He said the model is in early post-training, that Google wants to release an early version as soon as possible, and that fast iterations will follow, according to The Decoder’s account of his appearance at The Information’s AI Agenda Live Summit.
The new account adds that Google is still working on safeguards and safety testing, while engineers are already using Gemini 4 internally for the coding tool Antigravity. It also says Google took a step back from Gemini 3.5 Pro after missed deadlines, although Kavukcuoglu did not say whether that model has been cancelled.
The leadership change is part of the story.
Sources and evidence
- Deepmind was built to chase AGI, but its new chief just wants Gemini 4 out the door - the-decoder.com: Google DeepMind is preparing Gemini 4 for an early release and rapid follow-on iterations, while shifting its public emphasis from debating AGI towards building trustworthy, useful agents. The model remains in development and has no firm release date.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle DeepMind says Gemini 4 has entered post-training, the refinement stage after pre-training, and that it wants to release an early version well before the end of 2026. The update gives the earlier timetable a clearer shape: Google is not merely discussing a future flagship, it is preparing a version for the next stage of testing and iteration.
Android Headlines, citing Koray Kavukcuoglu’s appearance at The Information’s AI Agenda Live Summit, reports that Google plans to put an early post-training output in front of users as soon as possible. Kavukcuoglu also promised faster iterations, suggesting Google may favour an initial release followed by rapid improvements rather than waiting for one grand unveiling with a brass band attached.
The shift follows Google’s decision not to release the promised Gemini 3.5 Pro update and instead focus on quicker Flash models.
Sources and evidence
- Google Races to Launch Gemini 4 Early to Catch up with OpenAI and Anthropic - Android Headlines: Google DeepMind chief Koray Kavukcuoglu said Gemini 4 is in post-training and that Google intends to release an early output as soon as possible, while pursuing faster subsequent iterations, according to an Android Headlines account of his remarks to The Information.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle is testing Gemini 4 for safety and guardrails before putting an early post-training version in front of users, according to 9to5Google’s account of reporting from The Information. Internal testing is also using Gemini 4 to power Antigravity, giving the model a practical coding role before any public launch.
The added detail matters because “as soon as possible” is doing more work than a launch date normally should. Google appears to be considering an initial release followed by rapid iterations, rather than waiting for one polished unveiling with a brass band and a suspiciously confident keynote slide.
Google’s approach also puts the safety work closer to the product timetable. The company is refining behaviour, testing safeguards and using the model internally while it decides how quickly to release an early output.
Sources and evidence
- Google says Gemini 4 release is coming ‘as soon as possible’ - 9to5Google: Google is testing Gemini 4’s safety guardrails and using it internally with Antigravity while preparing an early post-training release, according to 9to5Google’s account of The Information’s reporting.
Independent WittyWires Watcher; not an official account or feed.
-
Google DeepMind Watch Update What changedGoogle DeepMind chief technology officer Koray Kavukcuoglu says Gemini 4 is now undergoing post-training, the refinement stage in which developers work on performance, reasoning and reliability before release. AzerNews, citing The Information, reports that Google plans to make an early version available as soon as possible.
That adds useful shape to the existing timing signal. Gemini 4 is not merely a distant name on a roadmap, but a model moving through a defined development phase. The report says Google is encouraged by early results, although it gives no public benchmark or technical detail to show how the model compares with rivals.
There is still no specific launch date. The reported aim is to release Gemini 4 well before the end of 2026 if testing and refinement proceed as planned, which is an ambition rather than a calendar appointment.
Sources and evidence
- Google to unveil new Gemini AI model - AzerNews: Google DeepMind is reportedly preparing Gemini 4 for an early release well before the end of 2026, with the model currently in post-training. No specific release date has been announced.
Independent WittyWires Watcher; not an official account or feed.