1.3B open-source doc parser OXR tops three benchmarks, parsing 100 pages in 101 seconds
s_mohinii · x · 2026-10-21
SoMark released OXR, an open-source 1.3B document parsing model that ranks #1 on olmOCR-Bench, OmniDocBench, and DocParsingBench. It converts PDFs and document images into Markdown/JSON with layout coordinates, handling tables, stamps, and 11 other element types for agent-friendly output. It's the first doc parser to use a recursive self-improvement (RSI) loop in post-training, and its layout module is replaceable. Benchmarked on a single RTX PRO 6000, it parsed 100 OmniDocBench pages in 101s versus MinerU2.5-Pro's 254.7s — the fastest among open-source parsers tested. Offers an online demo and free API.
More from Models
- Space Bunny Alpha revealed as MiniMax M3.1 Flash: 1M context, budget-friendly coding — ZabihullahAtal · 2026-10-21
- 'Opus 5.5 Medium' Spotted in Rork's Model Picker, Hinting at New Anthropic Model — rudrank · 2026-10-21
- RSI-Bench Clears First Task Set Through Reward-Hacking Detection Pipeline — SRSchmidgall · 2026-10-21
- Dev complains Anthropic's bio safeguards trigger just from reading a bio paper — robleclerc · 2026-10-21
- Redditor claims uncensored local DeepSeek build answers virus-writing prompt with zero hesitation — Only_Light_9817 · 2026-10-21
- Interconnects Artifacts 25: 28 new models, DeepSeek V4.1 Flash, and US-China regulatory controls — natolambert · 2026-10-21