OpenAI Releases o3-mini in Response to DeepSeek, Makes it Free for All

OpenAI has announced the release of OpenAI o3-mini, a new AI model built to deliver cost-effective reasoning with improved efficiency and STEM capabilities. The model is available in ChatGPT and the API, replacing OpenAI o1-mini in the model selection.

“o3-mini is out! smart, fast model. available in ChatGPT and API. It can search the web, and it shows its thinking. Available to free-tier users! click the “reason” button,” said OpenAI chief Sam Altman in a post on X. 

The launch of o3-mini comes amid the rising popularity of DeepSeek-R1, an open-source model that outperforms OpenAI’s o1.

Interestingly, during a Reddit AMA,  Altman was asked whether OpenAI would consider releasing some model weights and publishing more research. In response, he acknowledged the ongoing discussions within the company, saying, “I personally think we have been on the wrong side of history here and need to figure out a different open-source strategy.” 

However, he also noted that not everyone at OpenAI shares this view and that it is “not our current highest priority.” He further praised the DeepSeek model, saying, “It’s a very good model! We will produce better models, but our lead will be smaller than in previous years.”

OpenAI announced that “o3-mini advances the boundaries of what small models can achieve, delivering exceptional STEM capabilities while maintaining low cost and reduced latency.” The model supports function calling, structured outputs, and developer messages, making it more adaptable for production use.

OpenAI o3-mini is available in the Chat Completions API, Assistants API, and Batch API for select developers in API usage tiers 3-5. ChatGPT Plus, Team, and Pro users can access the model starting today, with Enterprise access set for February. 

Free-tier users can also try the model by selecting the ‘Reason’ option in the ChatGPT interface.

The model introduces three reasoning effort modes—low, medium, and high—allowing users to balance speed and complexity. “This flexibility allows o3-mini to ‘think harder’ when tackling complex challenges or prioritise speed when latency is a concern,” OpenAI stated.

Performance evaluations indicate that o3-mini surpasses its predecessor, o1-mini, in key STEM areas. On the AIME 2024 math competition, the high-reasoning variant of o3-mini achieved 83.6% accuracy. In PhD-level science evaluations (GPQA Diamond), o3-mini (high) reached 77.0% accuracy, showing improvements over previous models. In competitive programming tasks on Codeforces, o3-mini (high) achieved an Elo rating of 2073, surpassing o1-mini.

Software engineering benchmarks also show gains. In the SWE-bench Verified evaluation, o3-mini (high) achieved a 48.9% accuracy, exceeding o1-mini’s performance. In LiveBench coding evaluations, the model outperformed o1-high at medium reasoning effort, reinforcing its efficiency in coding tasks.

Beyond STEM, o3-mini demonstrates superior general knowledge performance. Human preference evaluations showed testers preferred o3-mini’s responses over o1-mini’s 56% of the time, with a 39% reduction in major errors on complex questions.

OpenAI also emphasised improvements in speed and latency. “o3-mini delivers responses 24% faster than o1-mini, with an average response time of 7.7 seconds compared to 10.16 seconds,” the company stated. The model has an average 2500ms faster time to first token than o1-mini.

Safety remains a key focus. OpenAI highlighted the use of deliberative alignment to ensure safer responses. “o3-mini significantly surpasses GPT-4o on challenging safety and jailbreak evaluations,” OpenAI noted. The model underwent external red-teaming and rigorous safety testing before deployment.

With this release, OpenAI continues its effort to improve AI accessibility while reducing costs. “o3-mini continues our track record of driving down the cost of intelligence—reducing per-token pricing by 95% since launching GPT-4—while maintaining top-tier reasoning capabilities,” OpenAI stated.

The company is adding search features to its reasoning models. Paid users can choose o3-mini-high for better reasoning, while Pro users get unlimited access to both o3-mini and o3-mini-high.