DeepSeek has released two "reasoning-first" AI models for software agents ‒ programs that can perform tasks autonomously ‒ according to an announcement on the Chinese startup's official WeChat social media account on Monday December 1.

The smartphone apps DeepSeek page is seen on a smartphone screen in Beijing, January 28, 2025. (Photo: AP)

"Reasoning-first" refers to AI systems optimized to break down problems step-by-step rather than simply predicting the next likely output.

The standard model, DeepSeek-V3.2, is now available on the company's website, mobile app and application programming interface (API), where developers can plug the model into their own software.

DeepSeek said the model balances strong reasoning with controlled output length, making it suitable for everyday uses such as answering questions and performing general automated tasks.

A second version, DeepSeek-V3.2-Speciale, is available only through the API for research and community testing.

The company said this variant is designed to push the limits of reasoning in open-source models whose all the underlying code is publicly accessible.

DeepSeek was founded in Hangzhou in 2023 and incubated by High-Flyer, a Chinese quantitative hedge fund.

The company's team is led by Liang Wenfeng and includes graduates of top Chinese universities and technical experts from international institutions.

In January, DeepSeek released its reasoning-focused model DeepSeek-R1, which it said matched the performance of OpenAI's o1 model at far lower cost.

The release earned DeepSeek a reputation as a fast-rising competitor in the large language model industry.