Overview
Deep Review is an agentic review system that Google Research released publicly on October 14, 2026, built to give rigorous technical verification on complex documents. It started with computer science papers and now covers knowledge documents in fields such as law and finance, and can also work through Google Docs and presentations.
It addresses a specific imbalance: content generation has sped up while human verification capacity has stayed roughly fixed, and long manuscripts or complex agreements still need specialist knowledge to check, which creates a validation bottleneck. Deep Review sits at that bottleneck, working as an iterative improvement tool during drafting and as a way to verify process, claims and evaluation during review.
It is not built from scratch. It evolves Google’s Paper Assistant Tool, and over the past year Google partnered with four flagship computer science conferences, STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027, to give private pre-submission feedback to authors of more than 16,000 manuscripts who opted in. The pilots are documented in Towards Automating Scientific Review with Google’s Paper Assistant Tool.
Key Features
- Agentic technical verification: Rigorous feedback on complex documents, checking evidence, argument and conclusions
- Beyond papers: Covers knowledge documents in law and finance as well as computer science, plus Google Docs and presentations
- Drafting and review: An iterative tool while drafting and a verification pass on process, claims and evaluation during review
- Two free reviews a week: Every user receives two complimentary document reviews per week
- Bring your Gemini API key: High-volume users can connect their own Gemini API key for higher limits
- Deep Review API: Programmatic access for embedding review into your own workflow
- PDF and plain text: PDFs and plain text at launch, with multi-file input coming soon
- Proven across four conferences: Pre-submission feedback on more than 16,000 manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027
Use Cases
- Researchers stress-testing a manuscript before conference submission
- Legal and finance professionals checking claims and their basis in long documents
- Teams iterating technical proposals and bids during drafting
- Reviewers triaging what needs a close look when time is short
- Embedding document review into an internal pipeline through the API
Pros
- Run across more than 16,000 manuscripts at four flagship conferences, not a lab demo
- Coverage reaches law and finance documents as well as papers
- Usable in both drafting and review stages without switching tools
- Two free reviews a week mean you can try it before paying
- An API for wiring review into existing document workflows
Pricing
Every user receives two complimentary document reviews each week. Connecting a Gemini API key raises usage limits and unlocks programmatic access through the Deep Review API, with API calls billed under Gemini pricing. Exact quotas and prices follow the official pages.
Summary
Deep Review is the agentic review system Google Research released publicly on October 14, 2026, built for rigorous technical verification of complex documents. It grew out of the Paper Assistant Tool, which spent the past year working with STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027 on more than 16,000 manuscripts, and now extends from computer science papers to knowledge documents in law and finance as well as Google Docs and presentations. Access comes in two tiers: two free document reviews per user each week, and higher limits plus programmatic use through the Deep Review API by connecting a Gemini API key. Input starts with PDF and plain text. It suits a verification pass before submission or delivery, and repeated iteration while drafting; for spelling or formatting, this level of review is more than the task needs.
Version History
- Deep Review goes public after piloting on 16,000 manuscripts across four flagship conferences (2026-10-14): On October 14, 2026, Google Research announced the public release of Deep Review. It is the evolution of the Paper Assistant Tool, and over the past year Google partnered with four flagship computer science conferences, STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027, to give private pre-submission feedback to authors of more than 16,000 manuscripts who opted in, as documented in Towards Automating Scientific Review with Google’s Paper Assistant Tool. The public release expands the system from deep mathematical and scientific peer review to rigorous verification across knowledge documents including legal and finance material, as well as Google Docs and presentations, and it works both as an iterative improvement tool during drafting and as verification of process, claims and evaluation during review. Access is tiered: every user receives two complimentary document reviews per week, and users with high-volume workflows can connect a Gemini API key for higher limits and programmatic integration via the Deep Review API. Input formats start with PDF and plain text, with multi-file input coming soon.