This page outlines operational boundaries, rate limits, weekly quotas, and evaluation topology tiers for the Deep Review API.
Document Boundaries
| Parameter | Limit | Details |
|---|---|---|
| Maximum document size | 20 MB | Enforced on both direct media streaming uploads and Google Drive imports. |
| Maximum manuscript tokens | 300,000 tokens | Approximately ~200 dense pages. Enforced during document ingestion and tokenization. |
| Supported file formats | PDF, TXT, MD, TEX, DOCX, PPTX, HTML; Drive: PDF, Docs, Slides | Direct media uploads accept PDF (application/pdf), plain text (text/plain), Markdown (text/markdown, text/x-markdown), LaTeX (text/x-tex, application/x-tex), DOCX (application/vnd.openxmlformats-officedocument.wordprocessingml.document), PPTX (application/vnd.openxmlformats-officedocument.presentationml.presentation), and HTML/XHTML (text/html, application/xhtml+xml). Google Drive imports (driveSource) accept PDF, Google Docs, and Google Slides. |
| Maximum anchored comments | 300 comments | A single review returns at most 300 anchored critique points. |
| Maximum bounding boxes | 50 regions | Each anchored comment can target up to 50 bounding boxes on document pages. |
Rate Limits and Quotas
The Deep Review API enforces per-minute request rate limits and rolling free-tier quotas:
| Quota | Limit | Scope | Details |
|---|---|---|---|
| Create requests | 10 requests / min | Per user | Applies to CreateDocument and CreateReview. |
| Create requests | 1,000 requests / min | Per project | Global project-wide limit. |
| Read requests | 50 requests / min | Per user | Applies to GetDocument, ListDocuments, DeleteDocument, DownloadDocument, EstimateReviewCost, GetReview, ListReviews, GetUserQuota, and GetOperation. |
| Read requests | 1,000 requests / min | Per project | Global project-wide limit. |
| Free-tier reviews | 2 reviews / 7 days | Per user | Rolling 7-day quota. Bypassed when providing a geminiApiKey in runConfig. Inspect status using GET /v1/userQuota. |
Review Topologies
When creating a review, configure the depth and speed of the multi-agent
evaluators using the topology field in runConfig:
| Topology | Speed | Reviewer Agents | Recommended Use Case |
|---|---|---|---|
LOW |
~1–2 min | Single fast pass | Fast preliminary checks, formatting checks, and local testing. |
MID (Default) |
~3–5 min | Parallel specialized reviewers | Workshop and conference pre-submission sanity checks. |
HIGH |
~7–12 min | Multi-angle debate & synthesis | Comprehensive peer review with structured multi-agent debate and synthesis. |
EXTRA_HIGH |
~15–25 min | Exhaustive multi-persona panel | Deep verification for major archival submissions, book chapters, and journal reviews. |
Pagination
List endpoints (documents.list and documents.reviews.list) support
cursor-based pagination:
- Default page size: 50 items
- Maximum page size: 100 items (values greater than 100 are coerced to 100)
- Use the
nextPageTokenfrom the response as thepageTokenparameter in subsequent requests.