Skip to main content

Batch Queue for high-volume jobs

Batch Queue handles large async parse workloads that do not need immediate turnaround. Submit a /parse_async request with queue_priority: "batch" to use it. Batch jobs consume 20% fewer parse credits and are guaranteed to complete within 12 hours of submission.

Deep Extract V3

A new Deep Extract model, V3, is available in alpha and will become the default soon. It runs faster than the current model with comparable accuracy, and is the recommended path for complex extractions, including workloads still using the legacy array extract or spreadsheet agent. To try it, enable Deep Extract and set the Deep Extract model to v3 in your request settings:

Faster processing across workloads

Deep Split, spreadsheet parsing, and HTML processing are all faster. A new Deep Split model, V2, is available in alpha with faster processing on long documents.

New Deep Extract model

A new Deep Extract model is available. It processes documents faster than the previous Deep Extract model and improves extraction accuracy. To use it, enable Deep Extract and set the Deep Extract model to v2 in your request settings:

Automatic mode for agentic tables

Agentic table processing now uses automatic mode by default. Automatic mode selects how much enrichment each table needs, keeping accuracy high while lowering latency. To keep full enrichment on every table, set the table mode to max. See agentic table modes to learn more.

New default models

On June 21, 2026, new default models rolled out. The layout model updated to v3, with more accurate layout detection that parsing and downstream extraction rely on. The agentic table, text, and key-value models updated to v3 for better processing of tables, text, and key-value regions. Track current versions on the model versions page, and pin a version with version pinning if you need stable output.

Discounted batch processing

Batch Queue is available for larger parse workloads that do not need immediate turnaround, such as backfills, data dumps, and overnight runs. Batch Queue processes these jobs at a lower cost. See Batch Queue to learn more.

Faster PDF rendering in Studio

Studio’s document display is now faster and smoother, which makes viewing long document corpuses easier.

Retiring the legacy OCR option

The legacy OCR option in the Parse API is being retired for all users. Starting Monday, June 22, 2026, requests that set the OCR system to legacy are served by the default OCR system instead. In the v3 config, settings.ocr_system: legacy is now served by standard. In the v2 config, advanced_options.ocr_system: legacy is now served by highres. These default OCR systems are faster than the legacy option.

Deep Split

Deep Split brings an agent-based approach to the Split endpoint, improving segmentation on long documents.

Improved job execution

Job execution is more reliable, with lower latency on long documents and large jobs.

Native MCP server

A native Model Context Protocol (MCP) server lets agents call Reducto directly from tools like Claude and Cursor, using plain language to run document work on local files. See the MCP server page to learn more.

Redesigned Studio

Studio has a refreshed look and feel, including light and dark modes.

New extract model

A new extract model is available for early access before it becomes the default. It improves accuracy and recall, especially on long schemas and long-context documents. To try it, set the extract model to v2 in your request settings. Citations must be enabled to use it:
Track current versions on the model versions page, and pin a version with version pinning if you need stable output.

Classify endpoint

The new Classify endpoint sorts documents into categories you define. Provide categories and criteria, and Classify assigns each document before you parse or extract.

Deep Extract

Deep Extract is a new extraction option for long and complex documents. It uses a longer-running, self-verifying agent to improve accuracy on difficult extractions.

Smart Schema

Smart Schema helps you build and refine extract schemas. It suggests schema improvements and asks clarifying questions based on your document’s context to improve results.

Faster parsing

Base parsing is now faster, so most non-agentic parse requests return more quickly. Spreadsheet parsing is also faster and more accurate.