Author of "DeepSeek-OCR" in GitHub
How this journalist typically writes
Based on 3 scored articles
Deepseek-Ai as author
DeepSeek-V3.2 achieves GPT-5 comparable performance with superior computational efficiency through sparse attention mechanisms and scalable reinforcement learning, with the high-compute variant outperforming GPT-5 and matching Gemini-3.0-Pro on mathematical and programming reasoning benchmarks.
“Author of "DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models"”
DeepSeek-OCR achieves 97% OCR accuracy at 10x compression ratio and surpasses GOT-OCR2.0 using only 100 vision tokens, demonstrating feasibility of compressing long contexts via optical 2D mapping.
“Author of "DeepSeek-OCR: Contexts Optical Compression"”
DeepSeek releases DeepSeek-OCR2, an advanced optical character recognition model designed to investigate the role of vision encoders from an LLM-centric perspective, with official vLLM integration support.
“Author of "DeepSeek-OCR" in GitHub”