Open-source doc parsing model OXR hits SOTA with 1.3B params and RSI post-training
JaynitMakwana · x · 2026-10-23
SoMark released OXR (Optical Everything Recognition), an open-source document parsing model that converts PDFs and document images into agent-friendly Markdown or JSON with layout coordinates, aimed at RAG pipelines and agents. Trained from scratch at just 1.3B parameters, OXR uses a recursive self-improvement (RSI) loop in post-training: multiple rollouts flag wrong or disagreeing results, a general VLM explains the failures, explanations are clustered into recurring problem types, and new image-label pairs are generated automatically — no hand labeling. The repo is on GitHub (273 stars) with an online demo and free API.
Related event: Open-Source 1.3B Document Parsing Model OXR Tops Three Benchmarks(6 posts)→
More from coding & agent
- Jack Dorsey's open-source AI agent "Company OS" repo hits 14.4K GitHub stars — JarnoDuursma · 2026-10-23
- If LLMs Become the Universal Interface, Who Discovers MCP Services Nobody Knows About? — ThomasBuildLab · 2026-10-23
- One Sentence, 6 Minutes: Open-Source Buddy App Runs Claude Code On-Phone to Build Android Apps — 3amae404 · 2026-10-23
- General Reasoning's RLCR trains agent organizations with collective rewards — hierarchies emerge — daniel_mac8 · 2026-10-23
- Google Play now installs private self-built apps as AI Studio turns prompts into Android apps — aitrendz_xyz · 2026-10-23
- One accidental cache write on a long Fable thread torched 99% of the Claude session limit — Aryvyo · 2026-10-23