Google updates Deep Think and opens an API early-access route

Google updated Gemini 3 Deep Think on 12 February, positioning its specialised reasoning mode for science and engineering problems with incomplete data or no single obvious answer. The new version became available to Google AI Ultra subscribers in the Gemini app. Researchers, engineers and enterprises could also express interest in selected early access through the Gemini API.

The company described examples involving technical mathematics review, crystal-growth methods and physical component design. These are accounts supplied in Google’s announcement, not independently reproduced Capvise experiments. For a research team, the operational change was the possibility of testing the mode inside a programmatic workflow, rather than relying entirely on the consumer chat interface. The announcement did not make API access unrestricted.