Elamur Research⌕
☰
home / papers / 2026 / DeepSeek-OCR 2: Visual Causal Flow

DeepSeek-OCR 2: Visual Causal Flow

Paper · 2026 · Haoran Wei, Yaofeng Sun and Yukun Li

Summary

Most vision-language models pass image patches to the LLM in a fixed top-left-to-bottom-right order. This ignores the logical structure of complex pages, such as multi-column layouts, tables and formulas. DeepSeek-OCR 2 addresses this by redesigning the encoder. Its DeepEncoder V2 keeps the compact SAM-based tokenizer of DeepSeek-OCR but replaces the CLIP module with a small Qwen2-style language model.

A custom attention mask lets visual tokens see one another bidirectionally. An equal number of learnable "causal flow" queries attend causally, so they can progressively rearrange the visual content into a semantically sensible sequence. Only these queries reach the DeepSeek MoE decoder, which effectively stacks two 1D causal reasoners to approach 2D understanding.

The authors test the model on document parsing with a budget of 256–1120 visual tokens per image. It beats DeepSeek-OCR by 3.73% on OmniDocBench v1.5 and recovers reading order more accurately. The work matters for producing LLM pretraining data and suggests LLM-style encoders could unify several modalities.

Key points

  • Architecture:
    • The tokenizer is an 80M-parameter SAM-base plus two conv layers with 16× token compression and output dimension 896.
    • The encoder is Qwen2 500M, used in place of CLIP.
    • The decoder is DeepSeek-3B MoE with about 570M active parameters.
  • Token budget: The model uses 256 to 1120 visual tokens per image. This comes from one global view (256 tokens) plus 0–6 local crops of 144 tokens each. The upper limit equals Gemini-3 Pro's maximum budget.
  • Benchmark results:
    • The model gains 3.73% over DeepSeek-OCR on OmniDocBench v1.5.
    • Reading-order Edit Distance improves across all 9 document types.
    • Text recognition is weaker on newspapers, which the authors attribute to the token cap and only about 250k training samples.
  • Production results: Repetition rate falls from 6.25% to 4.17% on online user-log images and from 3.69% to 2.88% in batch PDF processing.
  • Availability: Code and model weights are released at github.com/deepseek-ai/DeepSeek-OCR-2.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 works

Cite

Cite as Haoran Wei, Yaofeng Sun and Yukun Li. 2026. DeepSeek-OCR 2: Visual Causal Flow. arXiv:2601.20552.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2601.20552 · arXiv 2601.20552

@misc{wei2026deepseek,
  title = {{DeepSeek-OCR 2: Visual Causal Flow}},
  author = {Haoran Wei, Yaofeng Sun and Yukun Li},
  year = {2026},
  doi = {10.48550/arxiv.2601.20552},
  eprint = {2601.20552},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2601.20552}
}

References

56 references, 25 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marker. URL https://github.com/datalab-to/marker . ↗
  2. Nanonets-ocr-s, 2025. URL https://huggingface.co/nanonets/Nanonets-OCR-s . ↗
  3. Ocrverse, 2025. URL https://github.com/DocTron-hub/OCRVerse . ↗
  4. Ocrflux, 2025. URL https://github.com/chatdoc-com/OCRFlux . ↗
  5. Adept. Fuyu-8b. https://huggingface.co/adept/fuyu-8b , 2023. ↗
  6. G. AI. Gemini 2.5-pro, 2025. URL https://gemini.google.com/ . ↗
  7. J. Bai, S. Bai, S. Yang, S. Wang, S. Tan, P. Wang, J. Lin, C. Zhou, and J. Zhou. Qwen-vl: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966, 2023.
  8. S. Bai, Y. Cai, R. Chen, et al. Qwen3-vl technical report. arXiv preprint arXiv:2511.21631, 2025. URL https://arxiv.org/abs/2511.21631 . ↗
  9. S. Bai, K. Chen, X. Liu, J. Wang, W. Ge, S. Song, K. Dang, P. Wang, S. Wang, J. Tang, H. Zhong, Y. Zhu, M. Yang, Z. Li, J. Wan, P. Wang, W. Ding, Z. Fu, Y. Xu, J. Ye, X. Zhang, T. Xie, Z. Cheng, H. Zhang, Z. Yang, H. Xu, and J. Lin. Qwen2.5-vl technical report. arXiv preprint arXiv:2502.13923, 2025.
  10. N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko. End-to-end object detection with transformers. In European conference on computer vision, pages 213–229. Springer, 2020.
  11. J. Chen, L. Kong, H. Wei, C. Liu, Z. Ge, L. Zhao, J. Sun, C. Han, and X. Zhang. Onechart: Purify the chart structural extraction via one auxiliary token. In Proceedings of the 32nd ACM International Conference on Multimedia, pages 147–155, 2024. find
  12. C. Cui, T. Sun, S. Liang, et al. Paddleocr-vl: Boosting multilingual document parsing via a 0.9 b ultra-compact vision-language model. arXiv preprint arXiv:2510.14528, 2025. — [arXiv:2510.14528](https://arxiv.org/abs/2510.14528) find
  13. C. Cui, T. Sun, M. Lin, T. Gao, Y. Zhang, J. Liu, X. Wang, Z. Zhang, C. Zhou, H. Liu, et al. Paddleocr 3.0 technical report. arXiv preprint arXiv:2507.05595, 2025. — [arXiv:2507.05595](https://arxiv.org/abs/2507.05595) find
  14. M. Dehghani, J. Djolonga, B. Mustafa, P. Padlewski, J. Heek, J. Gilmer, A. Steiner, M. Caron, R. Geirhos, I. Alabdulmohsin, et al. Patch n’ pack: Navit, a vision transformer for any aspect ratio and resolution. Advances in Neural Information Processing Systems, 36:3632–3656, 2023. find
  15. J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and sh…
  16. A. Dosovitskiy. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929, 2020.
  17. H. Feng, S. Wei, X. Fei, W. Shi, Y. Han, L. Liao, J. Lu, B. Wu, Q. Liu, C. Lin, et al. Dolphin: Document image parsing via heterogeneous anchor prompting. arXiv preprint arXiv:2505.14059, 2025. find
  18. K. He, X. Zhang, S. Ren, and J. Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016.
  19. A. Huang, C. Yao, C. Han, F. Wan, H. Guo, H. Lv, H. Zhou, J. Wang, J. Zhou, J. Sun, et al. Step3-vl-10b technical report. arXiv preprint arXiv:2601.09668, 2026.
  20. S. Iyer, X. V. Lin, R. Pasunuru, T. Mihaylov, D. Simig, P. Yu, K. Shuster, T. Wang, Q. Liu, P. S. Koura, et al. Opt-iml: Scaling language model instruction meta learning through the lens of generalization. arXiv preprint arXiv:2212.12017, 2022.
  21. A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, et al. Segment anything. arXiv preprint arXiv:2304.02643, 2023.
  22. J. Li, D. Li, S. Savarese, and S. Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, pages 19730–19742. PMLR, 2023.
  23. Z. Li, Y. Liu, Q. Liu, Z. Ma, Z. Zhang, S. Zhang, Z. Guo, J. Zhang, X. Wang, and X. Bai. Monkeyocr: Document parsing with a structure-recognition-relation triplet paradigm. arXiv preprint arXiv:2506.05218, 2025. find
  24. A. Liu, B. Feng, B. Wang, B. Wang, B. Liu, C. Zhao, C. Dengr, C. Ruan, D. Dai, D. Guo, et al. Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model. arXiv preprint arXiv:2405.04434, 2024. find
  25. A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan, et al. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437, 2024. find
  26. A. Liu, A. Mei, B. Lin, B. Xue, B. Wang, B. Xu, B. Wu, B. Zhang, C. Lin, C. Dong, et al. Deepseek-v3. 2: Pushing the frontier of open large language models. arXiv preprint arXiv:2512.02556, 2025.
  27. C. Liu, H. Wei, J. Chen, L. Kong, Z. Ge, Z. Zhu, L. Zhao, J. Sun, C. Han, and X. Zhang. Focus anywhere for fine-grained multi-page document understanding. arXiv preprint arXiv:2405.14295, 2024. — [arXiv:2405.14295](https://arxiv.org/abs/2405.14295) find
  28. F. Liu and H. Qiu. Context cascade compression: Exploring the upper limits of text compression. arXiv preprint arXiv:2511.15244, 2025. find
  29. F. Liu, H. Wei, W. Zhao, G. Li, J. Peng, and Z. Li. Wb-detr: Transformer-based detector without backbone. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 2979–2987, 2021. find
  30. Y. Liu, J. Gu, N. Goyal, X. Li, S. Edunov, M. Ghazvininejad, M. Lewis, and L. Zettlemoyer. Multilingual denoising pre-training for neural machine translation. Transactions of the Association for Computational Linguistics, 8:726–742, 2020. find
  31. Y. Liu, Z. Zhao, L. Tian, et al. Points-reader: Distillation-free adaptation of vision-language models for document conversion. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 1576–1601, November 2025. find
  32. I. Loshchilov and F. Hutter. Decoupled weight decay regularization. In ICLR, 2019. find
  33. OpenAI. Gpt-4 technical report, 2023.
  34. L. Ouyang, Y. Qu, H. Zhou, J. Zhu, R. Zhang, Q. Lin, B. Wang, Z. Zhao, M. Jiang, X. Zhao, et al. Omnidocbench: Benchmarking diverse pdf document parsing with comprehensive annotations. In Proceedings of the Computer Vision and Pattern Recognition Conference, pages 24838–24848, 2025. find
  35. Z. Pang, Z. Xie, Y. Man, and Y.-X. Wang. Frozen transformers in language models are effective visual encoder layers. arXiv preprint arXiv:2310.12973, 2023. — [arXiv:2310.12973](https://arxiv.org/abs/2310.12973) find
  36. J. Poznanski, A. Rangapur, J. Borchardt, J. Dunkelberger, R. Huff, D. Lin, C. Wilhelm, K. Lo, and L. Soldaini. olmocr: Unlocking trillions of tokens in pdfs with vision language models. arXiv preprint arXiv:2502.18443, 2025.
  37. A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning, pages 8748–8763. PMLR, 2021.
  38. J. Redmon and A. Farhadi. Yolo9000: better, faster, stronger. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 7263–7271, 2017. find
  39. Rednote. dots.ocr, 2025. URL https://github.com/rednote-hilab/dots.ocr . ↗
  40. S. Ren, K. He, R. Girshick, and J. Sun. Faster r-cnn: Towards real-time object detection with region proposal networks. In Advances in neural information processing systems, pages 91–99, 2015.
  41. B. Seed. Seed1. 8 model card: Towards generalized real-world agency, 2025. find
  42. J. Su, Y. Lu, S. Pan, B. Wen, and Y. Liu. Roformer: Enhanced transformer with rotary position embedding. arXiv preprint arXiv:2104.09864, 2021.
  43. C. Team. Chameleon: Mixed-modal early-fusion foundation models. arXiv preprint arXiv:2405.09818, 2024.
  44. G. Team, R. Anil, S. Borgeaud, J.-B. Alayrac, J. Yu, R. Soricut, J. Schalkwyk, A. M. Dai, A. Hauth, K. Millican, et al. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805, 2023.
  45. B. Wang, C. Xu, X. Zhao, L. Ouyang, F. Wu, Z. Zhao, R. Xu, K. Liu, Y. Qu, F. Shang, et al. Mineru: An open-source solution for precise document content extraction. arXiv preprint arXiv:2409.18839, 2024. find
  46. B. Wang, B. Wang, C. Wan, G. Huang, H. Hu, H. Jia, H. Nie, M. Li, N. Chen, S. Chen, et al. Step-3 is large yet affordable: Model-system co-design for cost-effective decoding. arXiv preprint arXiv:2507.19427, 2025.
  47. C. Wang, S. Chen, Y. Wu, Z. Zhang, L. Zhou, S. Liu, Z. Chen, Y. Liu, H. Wang, J. Li, et al. Neural codec language models are zero-shot text to speech synthesizers. arXiv preprint arXiv:2301.02111, 2023.
  48. P. Wang, S. Bai, S. Tan, S. Wang, Z. Fan, J. Bai, K. Chen, X. Liu, J. Wang, W. Ge, et al. Qwen2-vl: Enhancing vision-language model’s perception of the world at any resolution. arXiv preprint arXiv:2409.12191, 2024.
  49. W. Wang, Z. Gao, L. Gu, et al. Internvl3.5: Advancing open-source multimodal models in versatility, reasoning, and efficiency. arXiv preprint arXiv:2508.18265, 2025.
  50. H. Wei, L. Kong, J. Chen, L. Zhao, Z. Ge, J. Yang, J. Sun, C. Han, and X. Zhang. Vary: Scaling up the vision vocabulary for large vision-language model. In European Conference on Computer Vision, pages 408–424. Springer, 2024. find
  51. H. Wei, L. Kong, J. Chen, L. Zhao, Z. Ge, E. Yu, J. Sun, C. Han, and X. Zhang. Small language model meets with reinforced vision vocabulary. arXiv preprint arXiv:2401.12503, 2024. find
  52. H. Wei, C. Liu, J. Chen, J. Wang, L. Kong, Y. Xu, Z. Ge, L. Zhao, J. Sun, Y. Peng, et al. General ocr theory: Towards ocr-2.0 via a unified end-to-end model. arXiv preprint arXiv:2409.01704, 2024. — [arXiv:2409.01704](https://arxiv.org/abs/2409.01704) find
  53. H. Wei, Y. Yin, Y. Li, J. Wang, L. Zhao, J. Sun, Z. Ge, X. Zhang, and D. Jiang. Slow perception: Let’s perceive geometric figures step-by-step. arXiv preprint arXiv:2412.20631, 2024. — [arXiv:2412.20631](https://arxiv.org/abs/2412.20631) find
  54. H. Wei, Y. Sun, and Y. Li. Deepseek-ocr: Contexts optical compression. arXiv preprint arXiv:2510.18234, 2025.
  55. J. Zhu, W. Wang, Z. Chen, Z. Liu, S. Ye, L. Gu, H. Tian, Y. Duan, W. Su, J. Shao, et al. Internvl3: Exploring advanced training and test-time recipes for open-source multimodal models. arXiv preprint arXiv:2504.10479, 2025.
  56. X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai. Deformable detr: Deformable transformers for end-to-end object detection. arXiv preprint arXiv:2010.04159, 2020. — [arXiv:2010.04159](https://arxiv.org/abs/2010.04159) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wei-2026-deepseek-ocr-2-visual/