China AI startup DeepSeek has launched an AI model that is said to be low-cost, high-performance, and comparable to OpenAI's, attracting market attention and posing a threat to companies developing large AI models. In this context, OpenAI, the developer of ChatGPT, officially released the lightweight AI model o3-mini on the 1st, which is the newest and most cost-effective model in the OpenAI reasoning series.
According to OpenAI's introduction, o3-mini is the company's first small-scale reasoning model that supports features for high-demand developers, providing excellent STEM (science, technology, engineering, mathematics) capabilities, especially in science, mathematics, and coding, while maintaining the low cost and low latency of o1-mini.
Evaluations by expert test personnel show that compared to o1-mini, o3-mini's answers are more accurate, clearer, and have stronger reasoning abilities. Test personnel preferred o3-mini's responses 56% of the time and observed a 39% reduction in its main errors.
In terms of usage, developers can choose low, medium, or high modes according to the reasoning workload to optimize for their specific application cases, and this flexibility allows o3-mini to think more seriously when facing complex challenges or prioritize speed when there are latency issues.
In terms of pricing, OpenAI offers different plan options, with Pro users having unlimited access to o3-mini, Plus and Team users having three times the rate limit of o1-mini, and free users being able to try the Reason mode in ChatGPT.
It is worth mentioning that this is the first time OpenAI has opened a reasoning model to free users.
Altman Admits Closed-Source Strategy Was a Mistake
At the same time, OpenAI CEO Sam Altman stated in a Reddit AMA on Friday that the company has "been on the wrong side of history" when it comes to AI open-sourcing, hinting that OpenAI may face a major strategic shift amid intensifying competition from China and growing attention on high-performance open-source models.
Sam Altman mentioned that OpenAI is discussing the possibility of publicly releasing model weights and research:
Personally, I think we've been on the wrong side of history on this and we need to figure out a different open-source strategy, and not everyone at OpenAI agrees with this view, and it's not the top priority for the company right now.
This statement contrasts sharply with OpenAI's increasingly closed development direction in recent years, for which the company has been criticized, including by Tesla CEO Musk, who is currently suing OpenAI for betraying its original open-source mission.
Additionally, in the AMA, Sam Altman directly addressed the impact of DeepSeek, stating that DeepSeek has a very good model and that OpenAI will continue to develop better models, but OpenAI's lead advantage will not be as pronounced as in the past few years.
