--- title: "DeepSeek taps Alibaba open-source AI technology to boost OCR performance" type: "News" locale: "en" url: "https://longbridge.com/en/news/273929541.md" description: "Chinese AI start-up DeepSeek has launched an upgraded OCR model, DeepSeek-OCR 2, utilizing Alibaba Cloud's open-source Qwen2-0.5b to enhance performance. This update, which follows the original model's release just three months prior, demonstrates the growing influence of China's open-source ecosystem in AI development. The new model achieved a 3.73% performance improvement over its predecessor and has been open-sourced on Hugging Face. DeepSeek aims to refine its architecture for broader applications while addressing previous criticisms regarding its original model's performance." datetime: "2026-01-28T03:38:02.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/273929541.md) - [en](https://longbridge.com/en/news/273929541.md) - [zh-HK](https://longbridge.com/zh-HK/news/273929541.md) --- # DeepSeek taps Alibaba open-source AI technology to boost OCR performance Chinese artificial intelligence start-up DeepSeek on Tuesday unveiled an upgraded version of its optical character recognition (OCR) model, incorporating an Alibaba Cloud-developed open-source system to boost performance. The new model, DeepSeek-OCR 2, replaced a key component of its original architecture with Alibaba Cloud’s lightweight Qwen2-0.5b model, according to a research paper released by the company. The update, which comes just over three months after DeepSeek launched the first version of its OCR system, underscores the growing role of China’s open-source ecosystem in advancing domestic AI development. Alibaba Cloud is the artificial intelligence and cloud computing arm of Alibaba Group Holding, which owns the Post. In the original model, DeepSeek relied on Contrastive Language Image Pre-training (CLIP), a neural network framework developed by Microsoft-backed OpenAI in 2021 that links images with text descriptions. In OCR applications, CLIP helps systems identify and interpret text embedded in images. DeepSeek said that replacing CLIP with Alibaba’s Qwen2-0.5b enabled its OCR model to process documents in a way that mimicked how humans read, following “flexible yet semantically coherent scanning patterns driven by inherent logical structures”, according to the research. Benchmark tests showed the updated model delivered a 3.73 per cent performance improvement over its predecessor, which the company described as a meaningful gain on an already high accuracy base, it said. DeepSeek has open-sourced DeepSeek-OCR 2 on Hugging Face, a widely used open-source AI developer platform. The collaboration highlights how Chinese AI developers are increasingly drawing on one another’s open-source innovations to accelerate progress. Last year, Beijing-based start-up Moonshot AI launched its Kimi K2 system, which borrowed elements from DeepSeek’s V3 architecture while introducing significant redesigns, according to a company researcher. That launch reverberated through the global tech community, with some experts calling it another “DeepSeek moment” – a reference to the surprise impact of DeepSeek’s V3 and R1 model releases in early 2025. DeepSeek’s latest OCR update follows fresh academic scrutiny of its original approach. Researchers from China and Japan recently challenged the initial DeepSeek-OCR research, arguing that the model showed inconsistent performance under certain conditions. Their study found that the original system’s accuracy in visual question-answering tasks could fall to about 20 per cent when exposed to additional text intended to influence its reasoning, compared with roughly 90 per cent accuracy for standard AI models. DeepSeek said in the Tuesday research it would continue refining its OCR architecture for broader applications, while pushing “towards a more comprehensive vision of multimodal intelligence”. ### Related Stocks - [688328.CN](https://longbridge.com/en/quote/688328.CN.md) - [BABA.US](https://longbridge.com/en/quote/BABA.US.md) - [KBAB.US](https://longbridge.com/en/quote/KBAB.US.md) - [BABO.US](https://longbridge.com/en/quote/BABO.US.md) - [BABX.US](https://longbridge.com/en/quote/BABX.US.md) ## Related News & Research - [BARCLAYS CEO SEEING 'CREEPING' IMPACT FROM AI](https://longbridge.com/en/news/287072350.md) - [Citadel CEO Ken Griffin was a prominent AI skeptic. Now he says, 'AI is real.'](https://longbridge.com/en/news/286683665.md) - [SoundHoundAI stock analysis: Buy or sell this AI stock?](https://longbridge.com/en/news/286826155.md) - [Microsoft Stock Is an AI Bargain That Investors Are Missing](https://longbridge.com/en/news/286680636.md) - [Jim Cramer Says Nvidia Should Stay Inside China’s AI Boom, Not Walk Away](https://longbridge.com/en/news/286804523.md)