Primary use stated or inferred cautiously from official documentation.
B
BAIDU / baidu/Unlimited-OCR-2
Unlimited OCR 2
Baidu model for extracting content from multi-page documents, announced on October 27, 2026. SR-SWA retains working references within a finite window to reduce attention cost growth. Input consists of page images in order and output is structured text. PDFs require prior conversion into images. Repository creation the previous day does not replace the announcement date.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | baidu/Unlimited-OCR-2 |
|---|---|
| Model type | Vision and documents |
| Access model | Open weights |
| License | MIT |
| Deployment | Self-hosting with Transformers or SGLang |
| Release | 27 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Text · Image |
| Exit | Text |
| Context window | Not published |
| maximum output | Not published |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | Hugging Face · Transformers · SGLang |
Processing books and reports, recognizing tables, formulas, headings and cross-page relationships. Transformers supports single- and multi-page inference; SGLang adds serving with page windows. The card uses a working reference set of 512 tokens as an SR-SWA example, not as the document's total input limit.
The unlimited-horizon name does not guarantee infinite documents, constant memory for the entire service or perfect accuracy. The example configuration limits output per page and concurrency; these are not universal model maxima. Custom code and an adapted SGLang version are required: pin and review dependencies before executing trust_remote_code.
03 / CAPABILITIES
What it can do
Orientation
Processing books and reports, recognizing tables, formulas, headings and cross-page relationships. Transformers supports single- and multi-page inference; SGLang adds serving with page windows. The card uses a working reference set of 512 tokens as an SR-SWA example, not as the document's total input limit.
Context
Not published · Not published
Tools and integration
Not published
Access
Hugging Face
Input modalities
TextImage
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
The card declares MIT and the official repository publishes that license. We have not verified a specific commercial price for Unlimited OCR 2; cloud availability of the previous OCR model does not establish an endpoint for this version. The published test environment uses an 80 GB NVIDIA A800, without demonstrating a minimum requirement.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
The official chart reports overall scores of 84.6 out of 100 on MPDocBench and 77.1 out of 100 on MarathonDocBench. It distinguishes documents with dozens and hundreds of pages. The figures are transcribed from this version's official asset, not Unlimited-OCR 1. These are not independent Inferama measurements; comparisons require preserving datasets, window configuration and evaluation rules.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| MPDocBench | 84,6 / 100 | Overall score (0–100) | View source ↗ |
| MarathonDocBench | 77,1 / 100 | Overall score (0–100) | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Unlimited OCR 2
Version added to Inferama's verified catalog.
07 / SOURCES