AWESOME or Overhyped? OXR Document Parsing, Tested With My Own Ugly PDFs

Get GPUs Runpod: https://get.runpod.io/pe48 Get Hostinger: https://www.hostg.xyz/SHJiu Coupon Code "PROMPT" 10% discount OXR is an open-source document parsing model that claims #1 on three benchmarks. I ran it for real on a rented RTX 4090 and tested it on an exam paper, a merged-header table, handwriting, a real arXiv PDF and a bad receipt photo. It worked well, and it failed twice. Repo: https://github.com/SoMarkAI/OXR Weights: https://huggingface.co/SoMarkAI/OXR-1.0 Online demo: https://free-oxr.somark.ai RunPod (cheap, reliable GPUs): https://get.runpod.io/pe48 Chapters 0:00 The claim 0:39 What OXR says 1:46 The free API (429) 2:13 Can I run it on my laptop? 2:28 Renting a GPU on RunPod 3:46 Getting it running (4 dead ends) 4:50 Real tests: exam, table, paper, receipt 8:42 License and verdict 10:05 Cost and wrap-up What I measured (one RTX 4090, Secure cloud, $0.89/hr, ~16 min, about $0.24 by my arithmetic): 0.7 s to 13.1 s per document; one demo-page run of the A3 exam took 43.7 s and looped on the header area; my synthetic receipt came back with prices shifted one row and no total. The #1 benchmark ranking is the authors' claim; I did not verify it. Notes: the receipt is synthetic (made by me). The intro/outro talking avatar is AI-generated from my likeness. License is Apache 2.0 plus extra terms (commercial thresholds, competing hosted API restriction): read the LICENSE file. 🔗 Connect with me: Email → [email protected] Ko-fi → https://ko-fi.com/promptengineer Patreon → https://www.patreon.com/PromptEngineer975 Book a Call → https://calendly.com/prompt-engineer48/call GitHub → https://github.com/PromptEngineer48 Twitter/X → https://twitter.com/prompt48 Website → https://promptengineer48.com/