AI4Bharat's IndicOCR: 0.8B model handles 9+ Indic scripts, rivals 4x larger ones

prajdabre · x · 2026-08-26

India's AI4Bharat, with BodhanAI, is set to release its first OCR model, IndicOCR, on Hugging Face and Bodhan on 05.09. The 0.8B-parameter model supports 9+ scripts, handles handwritten notes and degraded archives, and reportedly holds its own against models 4-5x its size. It completes the team's suite after Indic-Translate (text) and Indic-Transcribe (speech).

Related event: AI4Bharat to Open-Source 0.8B-Param IndicOCR Model(2 posts)→

Original post →

More from Multimodal

Multimodal channel →