DeepSeek recently shook up the AI landscape with r1, an open-weight model that matched OpenAI o1 in performance while being more than 20 times cheaper to run. This move set a new benchmark for cost-effective AI, putting pressure on OpenAI to respond. And they did—with o3-mini. It’s fast, efficient, and particularly strong in STEM fields like math, coding, and science. But as impressive as it is, is OpenAI “open”?
OpenAI o3-mini builds on o1-mini, offering better accuracy, reduced latency, and new features like function calling, structured outputs, and developer messages. It also allows users to adjust reasoning effort—low, medium, or high—depending on the complexity of the task. Unlike o1, however, it doesn’t support vision.
From a performance perspective, o3-mini is a solid upgrade. In head-to-head testing, expert evaluators preferred o3-mini over o1-mini 56% of the time, while major errors dropped by 39% on tough real-world problems. Even with medium reasoning effort, it matches o1 in AIME and GPQA—some of the hardest AI benchmarks. On top of that, it’s 24% faster than o1-mini, cutting response times from 10.16 seconds to 7.7 seconds.
For ChatGPT users, this means significant changes. The model replaces o1-mini in ChatGPT Plus, Team, and Pro, where rate limits are tripling from 50 to 150 messages per day. Pro users even get unlimited access to o3-mini-high, a version with enhanced reasoning. Meanwhile, free-tier users can now test a reasoning model for the first time by selecting ‘Reason’ in the chat interface.
Security and AI alignment remain a focus. OpenAI trained o3-mini using deliberative alignment, where the model first reasons through safety policies before responding. According to internal tests, it even outperforms GPT-4o on safety and jailbreak challenges.
But what about openness? OpenAI continues pushing intelligence forward while driving costs down—they claim a 95% per-token reduction since GPT-4. However, unlike DeepSeek’s fully open-source r1, o3-mini remains closed.
For developers and AI enthusiasts, o3-mini is a step forward, but the industry is watching closely. As open-weight alternatives like DeepSeek r1 gain traction, the question isn’t just about performance anymore—it’s about who controls the future of AI.