← Back to feed

chandra

GitHub Repo Pretty sure · license cutoff suggests monetization ...
https://github.com/datalab-to/chandra

Legitimately solid OCR that handles math/multilingual/layout — the hosted API is the real product, but the OSS inference option has actual utility.

25%
50%
25%
Slop 25%Signal 50%Science 25%

Chandra is a working document intelligence model that actually solves a hard problem (structured OCR with layout preservation across 90+ languages). The benchmarks look credible—tops olmocr, handles handwriting and tables competently. Real signal: you can pip install and run it locally or hit their API. The slop: README is benchmark-heavy marketing, and the monetization structure (free for <$2M revenue) telegraphs that the open release is a lead gen funnel for the hosted API. No novel researc...

6078 stars Python 2026-03-18 169 days old

Become a MFer to rate — log in