Falcon OCR 1.5: a 270M model challenges OCR rivals 5-20x its size
HildeKuehne · x · 2026-09-11
Falcon OCR 1.5 updates the 270M early-fusion OCR model with better SFT and RL to reduce hallucinations and improve formatting. V1 crushed academic PDFs; v1.5 extends to historical scans, handwriting, receipts and scene text, competing with models 5-20x larger.
More from Models
- Claude users found ways around safeguards for bioweapons research — luisdans · 2026-09-11
- Thesys' Diffusion Language Model OUI-1, Built on Gemma, Trends on Hugging Face — thesysdev · 2026-09-11
- DeepSeek V4.1 Flash is an overthinker: fixes bugs in 50 seconds, then obsesses over visual testing — teortaxesTex · 2026-09-11
- Subscriber: same $200 now buys far less frontier model access than a month ago — StewartalsopIII · 2026-09-11
- Fine-tuning Qwen 3 4B on 100 zebra puzzles yields +31% on MATH-500 — TGSCrust · 2026-09-11
- Dev argues DeepSeek V4.1 could have used a non-causal encoder for stronger reasoning — mgostIH · 2026-09-11