Tasks and inputs
Sarvam Vision 2.1 is listed for Document question answering, Image-to-text, Document OCR. It accepts image, pdf and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Sarvam Vision 2.1 is Sarvam's 3B document-intelligence VLM for OCR, tables, forms and Indic handwriting in 22 Indian languages plus English.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative ₹0.50/page (Digitise) and ₹1.00/page (Extract), excl. GST.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
A land record from 1974, typed in Marathi, stamped twice and annotated by hand in the margin. A Tamil textbook with a three-level table. A Bengali KYC form filled in ballpoint. These are the documents Sarvam Vision 2.1 was tuned for, and they are exactly where general-purpose vision models tend to stumble.
At the core is a 3B-parameter state-space vision-language model. Around it sit two helpers: a semantic layout parser and a pointer network that decides reading order. The VLM can read a page on its own, but Sarvam found accuracy is higher with the harness doing layout first, so that is how Document AI runs it.
Version 2.1 arrived on 24 September 2026. It adds three things the original February 2026 release lacked:
It also targets the hallucinations and inconsistent outputs that early users flagged. Sarvam reports an olmOCR-Bench score of 87.3 and 87.39 on its own Indic OCR benchmark; both are publisher figures.
| Item | Detail |
|---|---|
| Model ID | sarvam-vision |
| Endpoint family | Document AI, /doc-ai/v1 (Digitise, Extract) |
| Languages | 22 Indian + English, input and output |
| Output | HTML, Markdown or JSON; tables preserved |
| Page cap | 10 pages per PDF job |
| File cap | 200 MB |
| Indicative price | ₹0.50/page Digitise, ₹1.00/page Extract |
Notice the ID has no version suffix. You cannot pin 2.1 explicitly; you get whatever Sarvam currently serves behind sarvam-vision. Teams with strict reproducibility needs should log outputs and re-run a golden set after any announced update.
Pricing has moved, too. Earlier in 2026 the per-page rate was ₹1.5; current list prices are lower, which matches Sarvam's statement that inference optimisations let it undercut its launch pricing. Confirm today's figure on your own plan before you budget.
It is not a chat model. There is no free-form Q&A endpoint here; you digitise a document or extract fields you define. For conversational reasoning over the extracted text, pipe the output to Sarvam 105B.
Split a 40-page archive into 10-page jobs and push it through Digitise. Then pick your three ugliest pages: a faded scan, a multi-column layout, a handwritten form. Compare field-level accuracy against what you get today. Old scans and multi-column pages are where any OCR stack shows its seams, so weight them heavily.
Other options for scanned documents are collected under document OCR models.
Compare the model on the work you need to complete.
Sarvam Vision 2.1 is listed for Document question answering, Image-to-text, Document OCR. It accepts image, pdf and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
OpenBMB
MiniCPM-OCR 1.0 is OpenBMB's open 1.3B document-parsing model for text, tables, formulas and charts, with a free API and on-device apps.
OpenBMB
MiniCPM-OCR Jev is OpenBMB's open 2B judge that scores whether an OCR result for a text, formula or table region is correct, from 0 to 1.
Tencent
Youtu-Parsing-Omni is Tencent Youtu's 5B open parser that turns documents, charts, images, audio and video into one structured JSON format.
lightonai
LightOnOCR-3-0.8B is LightOn's smallest open document model: page images in, markdown out, with optional bounding boxes and chart tables.
NVIDIA
Nemotron 3.5 Parse is NVIDIA's 0.631B open document parser that turns page images into Markdown, HTML or JSON with tables, charts and reading order.
Cohere
Cohere Parse (parse-v5.0) is a 2.3B vision-language document parser that turns PDFs, slides and scans into Markdown for RAG, at $1.50 per 1,000 pages.