Google has confirmed that its next flagship AI model, Gemini 4, is very close to a public release, with work already in the “post-training” phase. This step involves refining model behavior, conducting safety tests, and building in guardrails. Internal trials are already running, including deploying Gemini 4 within Google’s Antigravity project.
DeepMind Chief Koray Kavukcuoglu said at this week’s AI Agenda Live Summit that Google is planning to release an early version of Gemini 4 “as soon as possible,” signaling a faster pace of iteration compared to previous model rollouts. Though there’s no exact date yet, he expressed hope the release will occur well before the end of 2026.
Where Gemini 4 Fits in Google’s AI Timeline
The last major flagship model from Google was Gemini 3 Pro, launched in November 2025, followed by Gemini 3.1 Pro in February 2026. Since then, there have been several updates under the Gemini and Flash brands, with Gemini 3.8 Flash being the most recent. An anticipated model called Gemini 3.5 Pro was teased over the summer but never released; according to Kavukcuoglu, Google paused that effort to focus on improving Flash models instead, with Gemini 4 now the primary priority.
Earlier in the week, both Anthropic and OpenAI shipped major upgrades to their own leading AI models—moves which have fueled speculation that Google may lag behind in the AI arms race. Kavukcuoglu pushed back on that idea, expressing confidence in his team’s ability to remain at the “frontier” of AI innovation despite the rapid pace of external competition.
Focus on Trust and Safety—More Than Hype
Beyond speed, Google is putting emphasis on trustworthiness. Kavukcuoglu noted that the conversation around advanced AI isn’t just about capabilities, but also whether intelligent agents can be trusted. The post-training stage for Gemini 4 includes safety and guardrail work that is meant to ensure Gemini’s behavior aligns with expectations and societal norms.
The internal Antigravity testing—designed to stress-test the model’s real-world behavior—is part of that effort. Google hasn’t given details about what kinds of tasks Antigravity encompasses, but it’s clear that these tests are central to bringing Gemini 4 to readiness.
Google is increasingly aiming to strike a balance: moving fast to keep up with competitors, while investing in robust safety frameworks so AI models behave reliably. That’s especially critical with Gemini 4 thought to be a leap over its predecessors in both scale and capability.
What this means for users and developers: Gemini 4 could bring new features and performance across Google’s products, potentially powering tools like Gemini Live, Gmail, Keep, and others previously using Flash updates. Its arrival might mark a turning point in how Google delivers AI, from incremental updates to major model shifts.
Watch for announcements from Google in the coming months. If all goes well, an early version of Gemini 4 will drop well before 2027—and with it, a stronger signal of how Google plans to manage competition, innovation, and AI safety going forward.
