Google shuts its Deep Research Pro preview agent on 23 October

Google shuts its Deep Research Pro preview agent on 23 October

8 October 2026: Google will shut down the deep-research-pro-preview-12-2025 agent in its Gemini API on 23 October 2026, so any application still sending it requests needs to move to one of the newer Deep Research agents this week. The same release notes, posted on 8 October, retire two Flash model names by silently routing their traffic to newer models.

Not yet independently verified. This rests on Google's own changelog and deprecations page, which still show 23 October 2026 as the shutdown date. The date has passed and we have not tested the old agent name or seen an independent report. We will update this when it can be confirmed, and remove this note.

The Deep Research migration

Google's release notes name two replacements and say what each is for:

• deep-research-preview-04-2026, "designed for speed and efficiency, ideal to be streamed back to a client UI";
• deep-research-max-preview-04-2026, "maximum comprehensiveness for automated context gathering and synthesis".

The change is a one-line edit: the agent parameter in interactions.create requests is switched from the old name to one of the new ones. The work is in what comes after. Research agents return long, sourced reports, and a team that tuned prompts, length limits or downstream parsing to the December 2025 preview should run its own test set against the new agent before the old one goes dark, rather than discover the differences in production.

Two Flash names now point elsewhere

The same entry deprecates gemini-3.7-flash and gemini-3.5-flash differently. Neither endpoint is being switched off. Instead, requests to gemini-3.7-flash are now automatically routed to gemini-3.8-flash, and requests to gemini-3.5-flash to gemini-3.6-flash, and the old names can no longer be selected in Google AI Studio. Google asks developers to update their model strings and read the guides for the newer models.

Rerouting keeps applications running, but it also means the model behind a pinned name has changed without a deploy on the customer's side. Output style, latency, token use and cost per request can all shift, and any evaluation baseline or audit record that names the old model no longer describes what produced the output. Teams with regulated or tested workflows should record the switch and re-run their checks.

What else is on the deprecation list

Google's deprecations page, which tracks earliest shutdown dates, also lists gemini-3.1-flash-tts-preview for shutdown on 17 November 2026, with gemini-3.8-flash-tts or gemini-3.8-flash-lite-tts as replacements, and gemini-3.1-flash-lite for 7 May 2027. On 6 October Google also deprecated gemini-3.1-flash-image in favour of gemini-nano-banana-2.1, without a shutdown date. The page notes that its dates are the earliest a model might be retired and that exact dates come with advance notice.

Sources

• Google AI for Developers: Gemini API release notes (8 October 2026 entry), 8 October 2026
• Google AI for Developers: Gemini deprecations (read 2026-10-20), 20 October 2026