Gemini Deep Research Agent (deep-research-pro-preview-12-2025)
by Google
Announced
2026-10-08
Complete
2026-10-23
Refund window closes
—
Refund status
No refund applicable
Google's Gemini API changelog entry for October 8, 2026 deprecates the `deep-research-pro-preview-12-2025` agent and sets its shutdown for October 23, 2026, about two weeks of notice. The deprecations page lists it under managed preview agents with a release date of December 11, 2025 and `deep-research-preview-04-2026` as the recommended replacement. The changelog also names `deep-research-max-preview-04-2026` as a second option. The migration Google describes is one change: update the `agent` parameter in your `interactions.create` requests. Both replacements are still previews with no shutdown date announced.
Refund flow
- 1
Not applicable. The Deep Research agent is billed pay as you go on the underlying Gemini tokens and tool usage, with no pre-paid balance tied to the agent string, so there is no refund category to open.
- 2
If you are billed for deep-research-pro-preview-12-2025 runs after October 23, 2026, treat it as a billing error and raise it with Google Cloud billing support with the invoice line.
- 3
Re-budget with Google's own per task estimates, read from the Deep Research docs on October 16, 2026: deep-research-preview-04-2026 about $1.00 to $3.00 for a typical task (around 80 search queries, 250k input tokens with 50 to 70 percent cached, 60k output tokens), and deep-research-max-preview-04-2026 about $3.00 to $7.00 (up to around 160 search queries, 900k input tokens, 80k output tokens). Google labels these as estimates at preview rates that may change.
- 4
Google publishes no matching estimate for deep-research-pro-preview-12-2025 on that page, so run one real task on the new agent and compare its token and search counts against your own logs before switching batch jobs.
Migration path
Recommended successor
Gemini Deep Research (deep-research-preview-04-2026)
Why this is the closest fit
It is the replacement named on the deprecations page and first in the changelog entry. Google describes it as designed for speed and efficiency and suited to streaming back to a client UI. It runs through the same Interactions API, so the request shape stays the same and only the agent string changes. If your use was long unattended reports, Google points to deep-research-max-preview-04-2026, described as maximum comprehensiveness for automated context gathering and synthesis.
What differs from the original
Google released both 04-2026 versions on April 21, 2026 and its changelog lists four additions with that release: collaborative planning, visualization support, MCP server integration and File Search. Collaborative planning is off unless you set collaborative_planning true, in which case the agent returns a plan instead of a report, so leave it off to keep the old one shot behaviour. With no tools parameter the agent uses Google Search, URL Context and Code Execution by default. Limits on the docs page: the agent runs only through the Interactions API and not generate_content, background=True is required and needs store=True, the maximum research time is 60 minutes with most tasks finishing within 20, custom function calling tools are not supported (remote MCP servers are), and structured outputs are not supported.
Alternatives by use case
You stream research progress into a chat or app UI
Switch the agent string to deep-research-preview-04-2026, which Google positions for streaming to a client UI
You run long competitive or due diligence reports in the background
Test deep-research-max-preview-04-2026, and budget for Google's $3.00 to $7.00 per task estimate rather than the $1.00 to $3.00 of the standard agent
You need the agent to search your own documents
Pass a file_search tool with your File Search store names, which the 04-2026 agents support
You parse the report into a fixed JSON shape
Keep your own parsing step, because the docs say the Deep Research agent does not support structured outputs
What this tool meant
This was the first Deep Research agent string on the Gemini API, released December 11, 2025. Its shutdown is a short one: the changelog entry is dated October 8 and the cutoff is October 23, and the replacements are previews too, so anyone building on Deep Research should expect the agent string to keep moving. The price is not a fixed rate card per task. It is the tokens and searches the agent decides to use, which is why Google gives ranges rather than one number.
Sources
- Google Gemini API changelog (October 8, 2026: deep-research-pro-preview-12-2025 deprecated, shutdown October 23, 2026; April 21, 2026: 04-2026 agent release)
- Google Gemini API deprecations page (managed agents table: release date, shutdown date, recommended replacement)
- Google docs: Gemini Deep Research agent (versions, tools, cost estimates, limitations)
- Google Gemini API pricing page (pricing for agents section)
- AVA graveyard: Gemini Live API preview November 17, 2026 shutdown
- AVA graveyard: gemini-omni-flash-preview October 22, 2026 shutdown
Spent credits on AI video that failed?
AVA automates the refund flow across every video provider
Free Chrome extension. Detects the failure mode, captures evidence, drafts the refund email with the technical term and Generation ID. Click send.