Changelog

• Parse v2 API updates
• Parse Jobs v2
• Extract v2 API
• Extract Jobs v2
• Service Tiers for Jobs
• New Playground URL
• Redesigned Usage page

​

Parse v2 API Updates

​

Parse API Changes

• Bounding boxes and page dimensions are now always in pixels. Previously, PDF coordinates were returned in points and image coordinates in pixels. The per-page unit field is removed from the response. A new dpi option (default 200, range 72 to 300) controls how PDF coordinates scale to pixels (pixels = PDF points × dpi / 72) and has no effect on image inputs. Each PDF page reports the dpi used, so you can re-render the page at a matching resolution and have the boxes line up.
• The options.elements field is renamed to options.blocks. The option that controls per-element captions and table format has a new name. For example, use options.blocks.table.format instead of options.elements.table.format. Update any requests that set these options.
• Table cells use a single table_cell type. The separate td and th element types are merged into one table_cell type. Header and data cells are no longer distinguished by type.
• The page break marker is now <!-- PAGE BREAK -->. The previous <!-- page --> comment is no longer emitted. Update any code that splits the Markdown on the page marker.
• Parse output ends with a document ID marker. The Markdown now appends <!-- doc_id=<job_id> --> as its final line. The Extract v2 API reads this automatically to link an extraction back to the parse job it came from.
• The grounding field is now a tree that mirrors structure. Previously, grounding was a flat object keyed by block id. It is now a document → page → block tree with the same order and ids as structure: each block node repeats its id and span and adds box and parts, a table node nests its cells’ grounding as children, and the per-block page field is removed (the page number lives on the page node). Walk the two trees together, or index the grounding nodes by id. See Grounding.
• The response metadata block has a new shape. The filename field is removed, version is renamed to model_version, and credit_usage is replaced by billing.total_credits. The new billing.service_tier field reports the service tier. See Metadata.
• Visual blocks have a new Markdown format. Figures render as HTML-style <figure type="TYPE">...</figure> elements containing transcribed text and generated <description> blocks, instead of image tags. Attestations emit a bracketed label line such as [SIGNED] followed by the transcribed text, and logos and scan codes render as transcribed text with short bracketed labels. See Figure and Attestation Labels.
• Math formulas are transcribed as LaTeX. Inline math is wrapped in $...$ and display math in $$...$$. The <sup> and <sub> tags now apply only to superscript and subscript text that is not part of a math formula.
• The v2 APIs are served only from api.ade.landing.ai. Code samples copied from the Playground early in the Preview (before June 3) point at an earlier host and path that no longer serve the v2 APIs. Update those requests to https://api.ade.landing.ai/v2/parse.
• Documents over the size cap now return HTTP 422. The Parse v2 API previously returned HTTP 413 for a document over the 50 MiB limit. It now returns a 422 error with the same message. Update any error handling keyed to 413.

​

Parse Jobs v2

• Submit and poll. Create a job, then check its status and retrieve the result when it finishes.
• Parse large documents. Jobs accept up to 6,000 pages and 1 GiB per PDF (50 MiB for images).
• Save output to your storage. Write the parsed output directly to a URL you control with output_save_url.

​

Extract v2 API

• Schema-driven extraction. Define the fields you want with a JSON schema and get consistent results across documents with varying layouts.
• Span-level grounding. Each extracted value includes [start, end) character offsets that map it back to its location in the source Markdown.
• Works on Markdown from any source. Extract from the Parse v2 API output, a third-party parser, or hand-authored Markdown.

​

Extract Jobs v2

​

Service Tiers for Jobs

• Standard tier. Consumes half the credits of the priority tier, with a slower turnaround. Best for cost-sensitive or non-urgent workloads.
• Priority tier. Consumes credits at the full rate, the same as synchronous processing, with a faster turnaround. Best for time-sensitive jobs.

​

New Playground URL

​

Redesigned Usage Page

• Daily and monthly consumption as a bar chart, so you can spot trends and spikes.
• Detailed per-job logs with new columns, including the job ID, the endpoint called, the model used (such as or ), and the mode (Sync or Async), alongside the credits each job used.

​

HTML Tables by Default

​

Parse Options in the Playground

​

Preview of DPT-3 and the New Parse API

​

Highlights

• Designed for agent systems. Stable element ids, predictable hierarchy, and Markdown span pointers so AI agents can map results back to source.
• Hierarchical layout model. The structured output is a nested hierarchy: pages, elements on each page, lines within text elements, and cells within tables.
• Line-level visual grounding. Bounding boxes per element, plus per-line boxes for text elements.
• Cleaner, more standardized Markdown. Standard Markdown image syntax for visuals, structured attestations and scan codes, standardized checkboxes, and figure subtype labels.
• New endpoint and response shape. Calls go to the new API, with a hierarchical structure tree alongside a flat grounding map keyed by element id. See the Migration Guide.

​

Get Started

• Overview
• Quickstart
• Migration Guide from DPT-2