Launches
Datalab Drops a 1M-Parameter OCR Model, Claims 49 Pages a Second on CPU
Developing 4 sources
I
Indra OCR
Model Oct 13, 2026
Datalab's downloadable CPU-focused OCR package combines a small recognition model with document layout and reading-order output.
Visit primary link
3 sources attached
What happened lately
Indra OCR timeline
Datalab announced Indra OCR on October 13 as a CPU-focused package for turning scanned images and PDFs into text, Markdown or JSON with line boxes, layout labels and reading order. Its main recognition model has about one million parameters; the package includes other components. Datalab reports 49.4 pages per second on a 32-core CPU in its own benchmark, not a universal speed guarantee. The code is under Apache 2.0, while the downloadable model card lists an OpenRAIL license. Datalab says handwriting and complex tables remain weak spots.