dots.mocr
dots.mocr is a 3B-parameter multilingual document parsing and image-to-text model licensed under MIT. It targets OCR, document layout analysis, table extraction, formula recognition, and SVG code generation from structured graphics. The model reports strong performance on standard benchmarks (olmOCR, OmniDocBench, XDocParse) relative to comparable-sized peers, though it remains below the largest generalist models (Gemini 3 Pro, Qwen3-VL-235B). It supports English and Chinese and is gated=false on HuggingFace.
Key facts
Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.
| Field | Value |
|---|---|
| Developer | rednote-hilab |
| Parameters | 3B |
| Context window | Unknown |
| License | mit — OSI-approved |
| Modality / task | image-text-to-text |
| Gated on HuggingFace | No |
| Downloads | 395.7k |
| Likes | 140 |
| Last updated | 2026-07-04 |
| Source | rednote-hilab/dots.mocr |
What dots.mocr is
3039M-parameter vision-language model with image-text-to-text pipeline. Capabilities include document OCR, semantic understanding, grounding, interactive dialogue, and direct SVG rendering from graphics. Evaluated on olmOCR-Bench (83.9%), OmniDocBench v1.5, XDocParse, and pdf-parse-bench. A specialized variant (dots.mocr-svg) is available for SVG-specific tasks. Context length is unknown. Model uses safetensors format and custom code; requires HuggingFace transformers for inference.
Run dots.mocr locally
Load the open weights with 🤗 Transformers and generate — the same model, self-hosted.
from transformers import pipelinepipe = pipeline("text-generation", model="rednote-hilab/dots.mocr")out = pipe("Explain retrieval-augmented generation in one sentence.", max_new_tokens=128)print(out[0]["generated_text"])Swap in vLLM or Ollama for production-grade serving. DEV.co can stand up the inference stack.
How you'd run it
A typical self-hosted path — open weights, an inference server, your application.
DEV.co builds each layer — from GPU infrastructure to the application.
Best use cases
Running & fine-tuning it
ESTIMATE: 3B parameters suggest 6–12 GB VRAM for inference in fp32; 3–6 GB in fp16 (bfloat16). Exact precision, quantization support (GPTQ, AWQ), and batching overhead are unknown and require testing. SVG generation variant may have different memory footprint.
Model card does not document LoRA, QLoRA, or full fine-tuning feasibility. Custom code is present, suggesting potential integration hooks, but fine-tuning infrastructure and training data requirements are not stated. Requires direct review of GitHub repository or contact with rednote-hilab.
When to avoid it — and what to weigh
- Real-time, ultra-low-latency inference required — 3B parameter model; exact latency/throughput characteristics are unknown. Benchmark code is from Gemini 3 Flash evaluation, not raw model speed measurements.
- Specialized domain OCR (medical, legal, code-heavy formats) — Evaluation focuses on general documents, tables, formulas, and graphics. No evidence of specialized fine-tuning for regulated or highly technical content.
- Languages beyond English and Chinese — Model explicitly supports only English and Chinese. Multilingual coverage is unclear; other languages are not documented.
- Guaranteed deterministic or audit-trail outputs — No information on output consistency, reproducibility modes, or provenance tracking for compliance-heavy workflows.
License & commercial use
MIT License. MIT is a permissive OSI-approved license permitting commercial use, modification, and distribution under attribution and liability disclaimers.
MIT license permits commercial use without restriction. No gating or restrictive terms stated in model card. However, verify that any third-party dependencies, training data, or custom code modules do not carry incompatible licenses. SVG parsing functionality and integration with proprietary pipelines should be validated for your use case. Recommend review of GitHub repository for any undisclosed dependencies.
DEV.co evaluation signals
Editorial assessment — not user reviews. Directional, with an explicit confidence level.
| Signal | Assessment |
|---|---|
| Maintenance | Active |
| Documentation | Adequate |
| License clarity | Clear |
| Deployment complexity | Moderate |
| DEV.co fit | Good |
| Assessment confidence | High |
Model uses custom code, requiring careful review before production deployment. No security audit, adversarial robustness, or prompt-injection testing is documented. SVG generation from untrusted images may pose injection risks if output is not sanitized before rendering. Input validation and output escaping should be implemented. No information on data retention or privacy handling in the live demo.
Alternatives to consider
PaddleOCR-VL-1.5
Smaller, similar benchmark performance on OmniDocBench (997.9 Elo vs. 1059.0), broader community support in Chinese-speaking regions, and integrated paddlepaddle ecosystem. Trade-off: fewer SVG capabilities and slightly lower average Elo (920.5 vs. 1124.7).
Gemini 3 Pro
Highest benchmark performance (1210.7 average Elo, 0.066 TextEdit error), supports more languages and modalities. Trade-off: API-only (no local inference), cost per request, privacy concerns, and vendor lock-in.
Qwen3-VL-235B-A22B-Instruct
Larger generalist model, strongest multilingual and multi-modal understanding, strong benchmarks. Trade-off: massive memory footprint (235B parameters), substantially higher inference cost, overkill for document-only tasks.
Ship dots.mocr with senior software developers
Start with the live demo at dotsocr.xiaohongshu.com, review the paper (arXiv:2603.13032), or clone the GitHub repository. For enterprise integration, custom fine-tuning, or deployment consulting, contact rednote-hilab or a Devco AI engineer to assess hardware requirements and integration complexity.
Talk to DEV.coRelated open-source tools
Surfaced by semantic similarity across the DEV.co open-source index.
Related on DEV.co
Explore the category and the services that help you build with it.
dots.mocr FAQ
Can I use dots.mocr in a commercial product?
What GPU/hardware do I need to run this locally?
Does dots.mocr support languages other than English and Chinese?
How do I fine-tune or customize dots.mocr for my documents?
Software development & web development with DEV.co
DEV.co helps companies turn open-source tools like dots.mocr into production software. Our software development services cover the full lifecycle — architecture, web development, integration, and maintenance — delivered by software developers and web developers who ship. Engage our software development agency to implement or customize it for your open-source llms stack.
Deploy dots.mocr for Your Document Automation Pipeline
Start with the live demo at dotsocr.xiaohongshu.com, review the paper (arXiv:2603.13032), or clone the GitHub repository. For enterprise integration, custom fine-tuning, or deployment consulting, contact rednote-hilab or a Devco AI engineer to assess hardware requirements and integration complexity.