OpenAI has begun rolling out o3-mini-high as the default reasoning model for ChatGPT’s free tier, replacing GPT-4o mini in the standard response pipeline. The change, confirmed by OpenAI in a post on X (formerly Twitter), represents the largest capability upgrade to the free version of ChatGPT since GPT-4o was briefly made available to non-subscribers in May 2024.
o3-mini-high is the highest-performance configuration of OpenAI’s o3-mini reasoning model family, which the company released in January 2026. In OpenAI’s internal benchmarks, o3-mini-high outperforms GPT-4o on coding tasks, mathematical reasoning, and multi-step problem solving, while remaining significantly faster than the full o3 model.
What changes for free users
Free ChatGPT users will now receive responses generated by o3-mini-high by default, without needing to manually select a model. The upgrade applies to standard text conversations; image generation via DALL-E and voice mode features remain restricted to ChatGPT Plus subscribers.
OpenAI has not announced whether the rollout includes usage caps for free users. Previously, free users accessing GPT-4o were subject to rate limits after a certain number of messages per hour, reverting to GPT-3.5 once the limit was hit. It is not yet clear whether o3-mini-high will operate under a similar cap structure or whether free users will receive uncapped access.
Context: competitive pressure from Google and Anthropic
The move comes amid intensifying competition in the consumer AI assistant market. Google upgraded Gemini’s free tier to Gemini 1.5 Pro in late 2025, and Anthropic has offered Claude 3.5 Sonnet — its most capable non-reasoning model at the time — to free users since mid-2025. OpenAI’s decision to make o3-mini-high the default for free users narrows the capability gap between paid and unpaid tiers across all three major platforms.
ChatGPT currently reports over 200 million weekly active users, the majority of whom use the free tier. The upgrade affects the largest segment of OpenAI’s user base.
What o3-mini-high is better at
The o3-mini model family was designed specifically for tasks requiring extended reasoning — problems where the model benefits from thinking through multiple steps before producing a response. Benchmarks published by OpenAI at the time of the January 2026 release showed o3-mini-high scoring 87.3% on the AIME 2024 mathematics competition and 69.1% on the SWE-bench Verified software engineering benchmark, compared to GPT-4o’s scores of 9.3% and 38.8% respectively on the same tests.
For everyday ChatGPT users, the practical difference is most noticeable in coding help, mathematical calculations, and tasks requiring the model to reason through a problem rather than pattern-match a response.
TechNexts will update this article as OpenAI releases further details about the rollout timeline and usage limits for free tier users.
