github datalab-to/chandra v0.2.0
Chandra OCR 2

7 hours ago

Announcing Chandra OCR 2 - a 4B parameter OCR model that scores 85.9% on the olmOCR benchmark (state of the art) and 77.8% on our internal 43-language multilingual benchmark. It is smaller and more accurate than Chandra 1 (9B) across every category.

Highlights

  • 4B parameters (down from 9B), 2x throughput improvement
  • 85.9% olmOCR overall (up from 83.1%)
  • 77.8% multilingual average across 43 languages (up from 69.4%), 72.7% across 90 languages
  • 2 pages/sec on H100 with 96 concurrent requests
  • 15+ layout block types with bounding boxes
  • Structured output for diagrams (Mermaid), charts, and images

olmOCR Benchmark

Category Chandra 1 Chandra 2 Change
ArXiv 82.2% 90.2% +8.0
Old Scans Math 80.3% 89.3% +9.0
Tables 88.0% 89.9% +1.9
Multi column 81.2% 83.5% +2.3
Overall 83.1% 85.9% +2.8

Multilingual

43-language averages: Chandra 2 77.8%, Chandra 1 69.4%, Gemini 2.5 Flash 67.6%, GPT-5 Mini 60.5%.

90-language averages: Chandra 2 72.7%, Gemini 2.5 Flash 60.8%.

Largest improvements in South Asian scripts (Bengali +27.2, Kannada +42.6, Malayalam +46.2, Tamil +26.9, Telugu +39.1). Full 90-language results in FULL_BENCHMARKS.md.

New Capabilities

  • Layout blocks: text, section-header, caption, footnote, table, form, list-group, image, figure, diagram, equation-block, code-block, chemical-block, bibliography, table-of-contents, page-header, page-footer, complex-block
  • Mermaid diagrams: Flowcharts and process diagrams converted to Mermaid format
  • Chart extraction: Structured data (values, categories, axis labels) from bar/line/pie charts
  • Image captioning: Captions generated from visual content and surrounding context
  • Chemistry detection: Molecular structure descriptions from chemical diagrams

Install

pip install chandra-ocr

# With vLLM (recommended)
chandra_vllm
chandra input.pdf ./output

# With HuggingFace
pip install chandra-ocr[hf]
chandra input.pdf ./output --method hf

Links

Don't miss a new chandra release

NewReleases is sending notifications on new releases.