Chinese artificial intelligence start-up DeepSeek has fully open-sourced an internally developed framework designed to run AI models efficiently, as a senior researcher at the company described the growing divide between open-source and closed-source AI as a battle between “light and darkness”.

The company on Tuesday released DeepInfer, an inference framework tailored to support its DeepSeek-V4.1-Flash model on both Nvidia’s SM100 Blackwell graphics processing units (GPUs) and Huawei Technologies’ Ascend 950 chips.

Inference refers to the serving and running of AI models, and frameworks such as DeepInfer help optimise that process.

Released in September, DeepSeek-V4.1-Flash introduced a new architecture that reduced the memory required to run the model while outperforming the company’s earlier V1 Pro model.

An update released on Saturday made it one of the world’s most powerful and cost-efficient open-weight models, according to third-party benchmarking firm Artificial Analysis.

As DeepSeek-V4.1-Flash is an open-weight model – allowing third-party cloud providers to serve it on their own infrastructure – open-sourcing DeepInfer would “help the community lower compute costs, reduce energy use and make AI more affordable”, the company said.

In a technical article published on Chinese question-and-answer platform Zhihu on Tuesday, DeepSeek suggested that a single computing cluster running its V4.1-Flash model on DeepInfer could generate revenue equivalent to US$38 per GPU per hour, based on off-peak pricing rates.

It added that the inference framework improved cost-efficiency by retaining previously processed information for more than 72 hours, reducing repeated computation. The company did not disclose its operating costs or profit margins. Open source vs closed source … A decisive battle between two futures, two destinies for humanityChen Deli, DeepSeek DeepSeek’s latest open-source release coincided with unusually combative comments from senior researcher Chen Deli.

In a social media post on Tuesday, Chen described open-source and closed-source AI as representing two fundamentally different futures for humanity – in some of the strongest language yet from a Chinese industry figure amid intensifying China-US competition in AI.

“However long they want to fight, we’ll fight. Until total victory.”Other DeepSeek staff have also weighed in on the company’s open-source push. Cui Tianyi, who leads DeepSeek’s Harness team, said in a social media post that the company hoped more businesses and organisations worldwide would “deploy DeepSeek models and profit from doing so”.

Chen’s remarks and DeepSeek’s announcement come as Chinese AI companies continue to roll out more open-weight model upgrades, challenging the dominance of leading American rivals such as Anthropic and OpenAI, which have championed closed-source AI.

In contrast to open-source AI, in which core technologies are released to the public and other commercial entities to be used and customised for their own use cases, closed-source AI remains proprietary technology owned and controlled exclusively by its developers.

Leading US industry figures, including Anthropic CEO Dario Amodei, have repeatedly argued that open-source AI posed risks to humanity, while also accusing Chinese companies of “stealing” American AI technology.

Despite such criticism, Chinese companies, including DeepSeek, Moonshot AI and Z.ai, have continued to release open-weight AI models.

That momentum has continued even as Beijing moves to tighten oversight of the sector. Last week, China proposed draft export controls covering 23 AI-related technologies. However, a government spokesperson said Beijing would continue to encourage open-source AI and promote its adoption worldwide, particularly across the Global South.