Gemini 4

Google confirms a new Gemini training run

Gemini 4 is Google DeepMind's next general-purpose model generation. A major new training run is confirmed, while its architecture, full multimodal feature set and public launch date remain undisclosed.

The most likely direction is one model spanning text, images, audio, video, search and tool use. The key question is not whether Gemini 4 exists, but whether it can materially improve long-horizon reasoning and autonomous work without increasing latency and cost. Specific capabilities remain open until an official specification appears.

Sources

Alphabet Q2 2026 remarksOfficial confirmation that Gemini 4 pre-training has started.Open primary source @Mr_SalioThe author infers a shift from Gemini 3.5 Pro to Gemini 4; Google has not confirmed that interpretation.Open primary source @imjustnewataiDiscusses possible model scale but provides no release timing evidence.Open primary source @VascoYapsA short recap of Alphabet's remarks, not independent confirmation.Open primary source