Chinese upstart DeepSeek released an update to its V3 large language model which it claims addresses real-world challenges, stepping up a battle for AI supremacy with rivals including OpenAI.

Released on developer platform Hugging Face and done so without a formal announcement, the V3-0324 update is for the company’s V3 LLM, which it released in December 2024.

According to Reuters, the latest model provides improvements to its predecessor in a range of areas including reasoning and coding capabilities, while setting benchmarks on Hugging Face for accuracy and efficiency. Its licence has also been changed to open-source from the Massachusetts Institute of Technology, making it more accessible to developers.

V3 was released before DeepSeek truly rose to prominence with the release of its R1 model in January, which helped propel it to the top of Apple App Store charts in the US.

Analysts argued R1 showed powerful AI models could be developed at a fraction of the cost of rival offerings, using less advanced chips.

Military use
Separately, Chinese media reported the country’s army has begun using DeepSeek’s open-source platforms for non-combat support functions, which could pave the way for deployment in high-risk situations.

Outlets reported DeepSeek’s models are being used in hospitals and solider training programmes, offering the military a controlled environment for experimentation.

If successful, Sam Bresnick, a research fellow at Georgetown University, told South China Morning Post that an advanced model like R1 could help the military with decision-making.

Autonomous driving solution provider Chongqing Landship Information Technology said in a whitepaper seen by Asia Times DeepSeek’s AI has potential to be used in the military, performing actions ranging from controlling unmanned vehicles to giving commands.