The Pulse
Google Says Gemini 4 Could Arrive Well Before Year-End
Google DeepMind chief Koray Kavukcuoglu says Gemini 4 is in early post-training and Google hopes to release an early version well before the end of 2026. He also explains why Google shifted focus from Gemini 3.5 Pro to smaller Flash models.
AI.info Team ·
Google DeepMind says Gemini 4 is already in early post-training, and the company hopes to release an early version “much earlier” than the end of 2026. Koray Kavukcuoglu, who leads Google DeepMind, described the model’s progress at The Information’s AI Agenda Live Summit in San Francisco on September 23. He did not name a launch date or say whether the first release would be generally available.
“Our intention is to, like—as soon as possible—to release an early post-training output because we see the results and we are excited,” Kavukcuoglu said in conversation with The Information’s editor-in-chief, Jessica Lessin. He said Google would then “continue the fast-paced iterations.” The comments put an early release, rather than a finished model’s arrival, at the center of the timeline.
Gemini 4 is still being refined
Post-training is the stage in which a base model is refined to improve how it behaves. Kavukcuoglu described Gemini 4 as being in the early days of that process, while also saying the team is encouraged by what it has seen so far. His remarks indicate an intended pace, not a firm commitment to a particular date.
Google is also developing safety guardrails and testing the model, according to The Information’s account of the interview. Kavukcuoglu said engineers are using Gemini 4 internally to run Antigravity, Google’s AI coding tool. The report did not specify what capabilities the model will offer at launch or which users would get access to an early version.
Google shifted away from Gemini 3.5 Pro
The next flagship would follow the Gemini 3 line, which launched late in 2025. Google released Gemini 3.1 in February and followed with smaller Flash models, but Gemini 3.5 Pro did not arrive after Google CEO Sundar Pichai said at the company’s May I/O conference that it would come in June.
Kavukcuoglu said Google “took a little bit of a step back” after Gemini 3 and 3.1, choosing to focus on Flash releases. “For us, what was important is to maximize the speed of learning at the time,” he said. He did not say Google had formally cancelled Gemini 3.5 Pro, but made clear that Gemini 4 is now the company’s focus.
A release window without a date
The timing matters for Google’s attempt to keep pace with rival model developers. The Information reported that Anthropic and OpenAI have released newer models since Gemini 3 that are more capable than Google’s current offerings. Kavukcuoglu said he trusts his team and is confident Google will remain at the frontier, but the interview offered no performance comparisons for Gemini 4.
Google has previously changed its stated release plans: Gemini 3.5 Pro missed the June window Pichai had given. Kavukcuoglu’s remarks now point to an early Gemini 4 release followed by continued updates, while leaving the exact schedule and the first version’s public availability unspecified. The Information’s full account of the interview is available here.