Legacy Parse is the previous Reducto pipeline. It runs OCR, layout detection, and table extraction as separate stages, and several of its features had to be turned on by hand. A request uses legacy Parse when it sets settings.model to "legacy", when it uses the V2 API, or when it omits settings.model on an account that is not on the r-1 plan.
Select the model
settings.model |
Pipeline |
|---|---|
"r-1" |
r-1. Recommended. |
"r-1-mini" |
r-1 mini, a smaller r-1 model for large, cost-sensitive workloads. |
"legacy" |
Legacy Parse. |
| not set | Legacy Parse until December 7, 2026, then r-1. |
The V2 API always uses legacy Parse. To use r-1, move the request to V3 with the V2 to V3 converter.
Legacy-only settings
r-1 ignores these settings. Remove them when you move a configuration to r-1.
| Setting | Legacy behavior | r-1 |
|---|---|---|
settings.ocr_system |
"standard" (default, multilingual) or "legacy" (Germanic languages only). |
Ignored. |
settings.extraction_mode |
"hybrid" (default) uses embedded PDF text with OCR. "ocr" uses OCR only. |
Ignored. |
enhance.summarize_figures |
Default true. Writes figure descriptions with a separate model. |
Ignored. r-1 describes figures by default. |
enhance.intelligent_ordering |
Default false. Fixes reading order with a separate model. |
Ignored. r-1 handles reading order. |
enhance.agentic entries with no prompt |
Run an extra model pass over text, tables, or figures to improve accuracy. | Remove. r-1 already gives this accuracy. Entries with a prompt are LLM enrichment. |
formatting.include: "highlight", "signatures", "ignore_watermarks" |
Turns on highlight detection, signature detection, or watermark removal. | Ignored. r-1 handles these by default. |
Not supported on r-1
A request that sets these with r-1 fails validation:
| Setting | Use instead |
|---|---|
formatting.table_output_format: "jsonbbox" |
"json", "html", or "md". For word and cell boxes, use Parse citations. |
track_offsets |
Parse citations, which return character offsets for each word. |
r-1 mini differences
r-1 mini follows the r-1 rules above, with two differences:
| Case | r-1 | r-1 mini |
|---|---|---|
| Spreadsheets | Supported. | Not supported on hosted deployments. Returns a 400 error. |
| On-prem deployments | Not available. | Available. |
Move a configuration to r-1
- Set
settings.modelto"r-1". - Remove every setting in the tables above.
- Keep chunking, table format, page range, filter blocks, spreadsheet, and the other processing settings. They work the same with r-1.
Legacy Parse is priced per standard and complex page. See Credit usage.