OmniParseBench: 16,288-test benchmark ranks OCR models on 2,937 pages
VikParuchuri · x · 2026-11-04
Datalab (Vik Paruchuri) released OmniParseBench, a document-parsing benchmark with 16,288 tests across 2,937 pages, including per-category breakdowns. It evaluates the recent wave of sub-1B OCR model releases.
Top of the leaderboard: Datalab's accurate mode leads at 94.1 ($7.50/1k pages), Gemini 3.8 Flash scores 92.9 ($37/1k pages), GPT-6.1 Sol 91.5, and Claude Opus 5.5 hits the best text score (93.9) but only 90.0 overall with 40 refusals. On the open side, chandra-ocr-2 (4.5B) reaches 90.1 and dots.mocr (3.0B) 87.9. Parsers differ notably across text/tables/layout; full results on GitHub and Hugging Face.
Related event: OmniParseBench Launches: Surya 2.5 Tops Small OCR Model Ranking(2 posts)→
More from Models
- Team says models trained across distributed GPUs work well, plans to share one soon — markjeffrey · 2026-11-04
- Researcher seeks help: hidden verifier prompts make his model eval-aware — nrehiew_ · 2026-11-04
- MiniMax H3 Lite and Highspeed 50% off until year-end, exclusive to Replicate and Cloudflare — ritakozlov · 2026-11-04
- Leaked chart shows OpenAI internal model scoring 0.875 on BenchCAD 2.0 vs GPT-6 Astra's 0.594 — ChrisGPT · 2026-11-04
- '6 Astra High' spotted in model picker, hinting at new Astra release — gethackteam · 2026-11-04
- DeepSeek V4.1 Flash reads raw weights: 3 of 4 RL checkpoints identified correctly — poppear · 2026-11-04