# DeepSeek-OCR 2: Visual Causal Flow

> DeepSeek AI replaces CLIP ViT with Qwen2-0.5B as the vision encoder and introduces causal flow queries that attend to document regions in semantic order. Achieves 91.09% on OmniDocBench v1.5 and outperforms Gemini-3 Pro at the same 1,120-token budget.

Canonical URL: https://kravhal.kcsatish.com/insights/week-09
Edition: Week 05 · March 2026
Tags: Deep Learning, Transformers, NLP
Reading time: 14 min read

---

This is a mirror of an article first published in the AI & Automation Chronicle.

Full text with the original formatting: https://chronicle.kcsatish.com/posts/week-09
Markdown of the original: https://chronicle.kcsatish.com/posts/week-09.md
Structured JSON of the original: https://chronicle.kcsatish.com/api/v1/posts/week-09.json

Cite the Chronicle as the publication of record for the research claims in this article.