Google Research announced worldwide public access to Deep Review on October 14, 2026. The AI document-review service is designed to examine complex material; Google says each user gets two complimentary reviews per week. Google Research describes the launch and the service’s purpose.

How Google says Deep Review reviews a document

Google describes a four-stage process. First, Deep Review divides a document into sections, which may overlap. It then allocates computing resources according to how much information each section contains. Specialized reasoning agents check sections in parallel while using the full document for context. Finally, the system brings the findings together, checks consistency across sections and performs checks against external sources.

Google presents the process as a way to check technical reasoning and evidence and to support drafting and review, while leaving higher-level judgments to people. The sequence explains the intended workflow; it is not a performance result.

Access, formats and API costs

Google’s October 14 launch post listed PDF and plain-text files as the initial input formats. The Deep Review page listed PDFs, text, Google Docs and Google Slides when viewed on October 15; the page also describes the service’s access and API terms.

For higher-volume, programmatic use, Google says users can provide their own Gemini API key and pay standard Gemini API charges, with no additional Deep Review fee. Its FAQ estimates about $3 for a 40-page document at Low effort with Gemini 3.8 Flash, or about $20 at Extra High with Gemini 4 Argon. Those are estimates; actual charges vary with the document and usage.

What the PAT benchmark measured

A preprint about Google’s earlier Paper Assistant Tool (PAT) reports a 34% improvement over zero-shot recall on SPOT, a benchmark for detecting mathematical errors. That result belongs to PAT and that benchmark task; it does not establish the public Deep Review service’s performance. The preprint reports the PAT result.

What Google says about uploaded documents

Google’s launch post says uploaded documents are stored for one year so users can view them in the interface, and are not used to train, fine-tune or improve Google models. The live Deep Review page describes a separate option: users can give permission for the research team to use data to improve Deep Review. Google’s live page sets out that opt-in.