MultimodalDeepSeek

DeepSeek-OCR

DeepSeek-OCR — Contexts Optical Compression

Why it matters

97% OCR precision at compression below 10×, about 60% at 20×. A shift in thinking: images as a medium for compressing context

Source — Primary sources first; secondary reports are labelled “Press”

Paperarxiv.orghttps://arxiv.org/abs/2510.18234

Added in Week 35, 2026 · See that week

Same week

Anthropic to use up to 1 million Google TPUs, with 1 GW+ online in 2026OfficialLangChain 1.0 and LangGraph 1.0 releasedOfficialHelion released in betaOfficialPyTorch releases MonarchOfficial

Nearby on the same axis

Sora 2 tightens deepfake guardrails on likeness rightsPressSora 2 swaps copyright opt-out for rightsholder opt-in with granular controlsOfficialOpenAI launches Sora 2 and a social appOfficialWorld Labs launches MarbleOfficial