Google Research opened Deep Review worldwide, offering two free weekly reviews and an API option for more intensive document-checking workflows.
Google Research announced the public release of Deep Review on October 14, 2026, expanding an AI system initially developed to help assess computer-science papers. The company says the tool is now available worldwide and can review complex documents in areas including research, legal work and finance.
Deep Review is designed to check reasoning and consistency across long documents, flag mathematical or factual claims for scrutiny, and suggest revisions. Google describes a four-stage process: a system divides a document into sections, allocates review effort according to complexity, sends agents to examine sections in parallel, and combines their findings with a final consistency and grounding pass. The system is intended to assist review, not replace expert judgment.
Google says each user can run two document reviews per week at no charge. Users with higher-volume needs can connect a Gemini API key for increased usage limits and programmatic access. Initial input formats are PDF and plain text; multi-file review is planned but was not yet available at launch. Google also says uploaded documents are not used to train or improve its models, but are stored for one year so users can access them in the interface.
The release follows pilots, conducted under the earlier name Paper Assistant Tool, with four computer-science conferences: STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027. Google reports that more than 16,000 opted-in manuscripts were covered across those deployments. Those trials provided pre-submission feedback and were separate from the conferences’ formal review processes.
Public access broadens the tool beyond conference pilots, giving researchers and other professionals a way to test complex documents before sharing or submitting them. But AI-generated critiques still need human checking: Google’s own research paper on the earlier system notes that it can falsely identify correct arguments as flawed. The company says it plans to add multi-file support, which could enable consistency checks across related documents and exhibits.