Chandra OCR 2

Name: Chandra OCR 2
Author: Datalab

Datalab · released 2026-03-16 · data checked 2026-11-06

Chandra OCR 2 is a 5.3B-parameter OCR and document-parsing model from Datalab (US), released 2026-03-16 under the OpenRAIL-M (modified) license. It scores 85.8 on olmOCR-Bench (#5 of 24). It needs about 11.6 GB of memory at 16-bit.

90+ languages; Markdown, HTML and JSON with layout. Datalab also makes the Marker and Surya tools.

OmniDocBench v1.6

—

Not reported

olmOCR-Bench

85.8

#5 of 24 models

Parameters

5.3B

OCR specialist

Memory to run

~11.6 GB

16-bit weights plus about 1 GB

Benchmarks

olmOCR-Bench Lab-reported85.8

Best: OXR 1.0 87.8

The tick marks the best score on each benchmark. See how to read the ranking.

Specifications

Lab: Datalab (US)
Type: OCR / document parsing specialist
Total parameters: 5.3B
License: OpenRAIL-M (modified)
Free for research, personal use and startups under $2M in funding or revenue, and not for competing with Datalab's API; otherwise a commercial license is needed.
Memory to run: ~11.6 GB at 16-bit
Released: 2026-03-16

Weights and usage on Hugging Face ↗

Closest alternatives

Infinity-Parser2 Flash (2.2B, 86.0 on olmOCR-Bench, Apache-2.0)
LightOnOCR-3 4B (4.5B, 86.3 on olmOCR-Bench, Apache-2.0)
LightOnOCR-3 1B (1B, 84.5 on olmOCR-Bench, Apache-2.0)
Infinity-Parser2 Pro (35.1B, 87.6 on olmOCR-Bench, Apache-2.0)

Sources

https://huggingface.co/datalab-to/chandra-ocr-2

Something wrong here? Report it.