OpenAI releases cheaper model in wake of DeepSeek

OpenAI rolled out a new, lightweight artificial intelligence model that it said is capable of humanlike reasoning and makes gains on efficiency — a product that will be closely watched following the release of an open-source model from China’s DeepSeek.

The model, released Friday, is called o3-mini and will be free to use. The company first announced plans for o3-mini in December at the same time it unveiled a more powerful version, o3. Both models are meant to answer complicated questions related to topics such as coding, math and science.

The timing is expected: OpenAI said last year that o3-mini would be released at the end of January and o3 would come shortly thereafter. But the release also coincides with an industry panic about the gains that DeepSeek appears to have made.

DeepSeek’s model, R1, rivals or outperforms leading US developers on a range of industry benchmarks, the company said. For example, it can complete mathematical tasks and answer questions with general knowledge. And, critically, it was built for a small fraction of the cost — teeing off a larger discussion about the price and value of such models. OpenAI is currently in talks to raise a massive funding round of $40 billion at as much as a $300 billion valuation.

OpenAI said Friday that o3-mini is now available via its ChatGPT chatbot to free users and paid ChatGPT Plus users, as well as those who pay for its Pro and Team services. The model is also available through the company’s application programming interface, or API. The company will offer it to enterprise users in a week, it said.