Chandra OCR 2
Chandra OCR 2 is Datalab's second open OCR model: smaller than Chandra 1, with better math, tables, layouts and support for 90+ languages.
A release of Chandra.
- Released
- March 2026
- Model ID
- datalab-to/chandra-ocr-2
- Size
- About 5.3B (Chandra 1 was 9B)
- License
- Code: Apache-2.0. Weights: modified OpenRAIL-M, free for research, personal use and startups under $2M in funding or revenue, and not for use in competition with Datalab's API
- Also searched as
- chandra ocr 2, chandra 2, chandra-ocr-2
What is new in Chandra OCR 2
- Scores 85.8 on olmOCR-Bench, up from 83.1 for Chandra 1.
- Averages 77.8% on Datalab's 43-language benchmark, which Datalab calls a 12% improvement over Chandra 1.
- Better math, tables and complex layouts, especially on wide documents.
- Better image captions.
- About 5.3B parameters against 9B for Chandra 1.
Chandra OCR 2 benchmarks
| Benchmark | Score | Note |
|---|---|---|
| olmOCR-Bench, overall | 85.8 ± 0.8 | Chandra 1: 83.1, olmOCR 2: 82.4 |
| 43-language benchmark, average | 77.8% | Datalab's own benchmark |
| 90-language benchmark, average | 72.7% | Gemini 2.5 Flash: 60.8% |
| Throughput | 1.44 pages/s | vLLM, one H100 80GB, 96 concurrent sequences |
Scores as published by the authors. We have not rerun them.
Use Chandra OCR 2 through an API
OCRRouter runs a different release of Chandra as datalab/chandra.
Chandra versions
- Chandra: all versions, price, speed and features on OCRRouter
- Chandra OCR 2 (March 2026), this page
Questions
Sources
Last updated .