Short answer: Google announced Gemini 3 Deep Think API access for select researchers, engineers, and enterprises through an early-access program. That announcement does not establish that Deep Think has a public API or that the early-access program is currently open. The available sources do not provide a verified Deep Think API model ID or request example. If you need a documented Google API model, Google’s Gemini 4 Argon guide includes a working request example for the separate model gemini-4-argon. (blog.google)
Is Gemini 3 Deep Think available through an API?
Google’s announcement said Deep Think would be available through the Gemini API to select researchers, engineers, and enterprises, with an option to express interest in early access. It described this as a restricted access route, not general public availability. The announcement does not, by itself, confirm whether the program is still accepting applicants. (blog.google)
A February 12, 2026 report by The Decoder also described a separate Vertex AI early-access API program. That report is useful historical context, but it does not provide current access terms or instructions for making a request.
Keep API access distinct from access in the Gemini app: app availability or a subscription does not establish API eligibility. The sources reviewed here do not verify a current Deep Think API model identifier, endpoint, authentication procedure, request format, API price, or production commitment. (blog.google)
A documented API alternative: Gemini 4 Argon
Google documents Gemini 4 Argon as a stable API model with the ID gemini-4-argon. The Gemini API release notes mark it generally available on October 13, 2026. Argon is a separate model—not a Deep Think API alias or a guarantee of equivalent behavior. (ai.google.dev)
Minimal REST request
Set GEMINI_API_KEY in your environment, then send a request to Google’s documented Interactions API endpoint. The endpoint, authentication header, model field, and input field below follow Google’s Argon quickstart: (ai.google.dev)
curl -X POST 'https://generativelanguage.googleapis.com/v1beta/interactions' \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H 'Content-Type: application/json' \
-d '{
"model": "gemini-4-argon",
"input": "Compare these two system designs and identify the main engineering trade-offs."
}'
Google’s quickstart shows the returned SDK interaction’s generated text accessed as interaction.output_text. That is an SDK property; use the documented response handling for the API surface and SDK you choose. The example calls Gemini 4 Argon, not Gemini 3 Deep Think. (ai.google.dev)
Google lists medium as Argon’s default thinking level and documents low, medium, and high as supported levels. Those are Argon API controls; they should not be presented as Deep Think API parameters. (ai.google.dev)
Pricing and production status
Google lists Argon as stable, and its release notes say it reached general availability on October 13, 2026. The published introductory rates are $2 per 1 million input tokens and $10 per 1 million output tokens through November 30, 2026. The same documentation lists $4 per 1 million input tokens and $20 per 1 million output tokens thereafter. These are Argon API rates, not Deep Think API pricing or a consumer subscription price. (ai.google.dev)
No current Deep Think-specific production commitment or API price is verified by the sources cited here. Do not use an Argon rate—or a Gemini app subscription price—to estimate Deep Think API costs.
If you specifically need Deep Think
Use Google’s official early-access announcement to check the documented access route, but confirm directly whether the program is currently accepting applicants and what terms apply before building around it. Until Google documents a Deep Think API identifier and request specification for your access route, avoid guessing a model name or treating an app setting as an API parameter. If you need an API call you can implement from public documentation, evaluate Gemini 4 Argon separately rather than assuming it is equivalent to Deep Think. (blog.google)
Sources
Gemini 3 Deep Think: AI model update designed for science
Google DeepMind upgrades Gemini 3 Deep Think for complex science and engineering tasks