Multimodal OCR: Parse Anything from Documents

Dimensions overall score 8.0

• The Multimodal OCR (MOCR) paradigm jointly parses text and graphics into unified textual representations, treating visual elements as first-class parsing targets.method 95%
  Evidencepartial

  We present Multimodal OCR (MOCR), a document parsing paradigm that jointly parses text and graphics into unified textual representations.

  Implicationpartial

  Explicitly stated in the abstract as the core contribution of the paper.

  Verificationpartial

  partial

• The dots.mocr model achieves a state-of-the-art score of 83.9 on the olmOCR Bench benchmark.result 90%
  Evidencepartial

  sets a new state of the art of 83.9 on olmOCR Bench.

  Implicationpartial

  Direct numeric result stated in the abstract with a specific benchmark score.

  Verificationpartial

  partial

• On document parsing benchmarks, dots.mocr ranks second only to Gemini 3 Pro on the OCR Arena Elo leaderboard and surpasses existing open-source systems.result 90%
  Evidencepartial

  On document parsing benchmarks, it ranks second only to Gemini 3 Pro on our OCR Arena Elo leaderboard, surpasses existing open-source document parsing systems

  Implicationpartial

  Direct comparative performance claim with a named competitor and category.

  Verificationpartial

  partial

• On structured graphics parsing, dots.mocr achieves higher reconstruction quality than Gemini 3 Pro across image-to-SVG benchmarks.result 90%
  Evidencepartial

  On structured graphics parsing, dots.mocr achieves higher reconstruction quality than Gemini 3 Pro across image-to-SVG benchmarks

  Implicationpartial

  Direct comparative performance claim against a named competitor on a specific task.

  Verificationpartial

  partial

• The method converts previously discarded graphics into reusable code-level supervision, unlocking multimodal supervision embedded in existing documents.technical 85%
  Evidencepartial

  it converts previously discarded graphics into reusable code-level supervision, unlocking multimodal supervision embedded in existing documents.

  Implicationpartial

  Explicitly stated as an advantage of the method in the abstract.

  Verificationpartial

  partial

• The system is trained on a comprehensive data engine built from PDFs, rendered webpages, and native SVG assets.method 85%
  Evidencepartial

  we build a comprehensive data engine from PDFs, rendered webpages, and native SVG assets

  Implicationpartial

  Directly stated in the abstract as part of the method's implementation.

  Verificationpartial

  partial

• The model architecture is a compact 3B-parameter model trained through staged pretraining and supervised fine-tuning.technical 85%
  Evidencepartial

  train a compact 3B-parameter model through staged pretraining and supervised fine-tuning.

  Implicationpartial

  Direct specification of model size and training approach in the abstract.

  Verificationpartial

  partial

• A potential limitation is the reliance on training datasets that may not cover all graphical elements seen in real-world documents.limitation 80%
  Evidencepartial

  the reliance on training datasets that may not cover all graphical elements seen in real-world documents.

  Implicationpartial

  Explicitly stated in the analysis excerpt as a caveat.

  Verificationpartial

  partial

Competitive landscape

A next-gen OCR system that parses documents into structured text and graphics for seamless integration and data retrieval.