Chinese AI developer DeepSeek and Peking University have jointly released DSpark, a new open-source tool aimed at making large language models run more efficiently.
According to Pandaily, the two partners are describing DSpark as a "major leap" in LLM inference efficiency. Inference is the stage where a trained AI model actually answers questions or generates text — the part that runs every time someone uses a chatbot, and a major driver of the cost and computing power behind today's AI services.
According to the South China Morning Post, DeepSeek says DSpark delivers faster AI at lower cost, while easing bottlenecks and reducing strain on chips. In other words, the tool is pitched as a way to squeeze more performance out of existing hardware rather than simply buying more of it.
By open-sourcing DSpark, the partners are making the code freely available for other developers and companies to use, inspect and build on — a strategy DeepSeek has leaned on before to widen the reach of its technology.
The two sources here are brief, and details such as benchmark figures, supported hardware and licensing terms were not specified in the items provided. What is clear is the framing: a collaboration between one of China's most closely watched AI labs and one of its top universities, targeting the expensive, compute-hungry side of running AI models.
Why it matters: with the cost and chip demands of running AI a central concern across the industry, a free tool that promises to cut those burdens could lower the barrier for more developers to deploy large language models.