OpenAI Launches o3 and o4-mini Reasoning Models Publicly

What Happened

OpenAI released two new reasoning-focused AI models, o3 and o4-mini, to the public on April 16, 2025, marking the company's broadest deployment yet of its "thinking" model architecture. The release makes o3, previously available only in limited research and API preview contexts, accessible to ChatGPT subscribers and developers through OpenAI's API.

Background

OpenAI's reasoning model line, which began with o1 in late 2024, is designed to spend additional compute time working through multi-step problems before returning a response. The approach differs from the company's GPT series, which generates answers more directly. The o-series models are intended for tasks requiring logical deduction, mathematics, coding, and scientific reasoning.

o3 was first demonstrated publicly in December 2024, when OpenAI reported benchmark results showing the model achieved high scores on several established evaluations, including the ARC-AGI benchmark, a test designed to measure general reasoning ability. At the time, the model was not released for general use. o4-mini, announced alongside o3, is a smaller and more cost-efficient variant intended for developers who require reasoning capabilities at lower latency and cost.

What Was Released

The public release covers two distinct models. o3 is positioned as OpenAI's most capable reasoning model to date and is available to ChatGPT Plus, Pro, and Team subscribers, as well as through the OpenAI API. o4-mini is available through the same channels and is priced lower per token, targeting use cases where speed and cost efficiency are priorities over maximum capability.

Both models support tool use, including web browsing, code execution via the Python interpreter, and image analysis, capabilities that were not fully available in earlier o-series releases. OpenAI has stated that o4-mini, despite being the smaller of the two models, outperforms the earlier o3-mini on several internal benchmarks.

According to OpenAI's published technical materials, o3 achieves strong results on the AIME 2024 mathematics competition benchmark and on SWE-bench Verified, a software engineering evaluation that tests a model's ability to resolve real GitHub issues. The company has not released a full technical report for either model at the time of the public launch.

Pricing and Access

For API users, OpenAI has set o3 pricing at $10 per million input tokens and $40 per million output tokens. o4-mini is priced at $1.10 per million input tokens and $4.40 per million output tokens. Both figures position the models within a competitive range relative to similar reasoning-oriented offerings from Anthropic and Google DeepMind, which have released their own thinking-capable models in the same period.

ChatGPT Free tier users do not have access to o3. OpenAI has not announced a timeline for expanding o3 access to free accounts.

Market Context

The release comes during an active period of competition in the AI reasoning model segment. Anthropic has made Claude 3.7 Sonnet available with extended thinking capabilities, while Google DeepMind has released the Gemini 2.5 Pro and Flash models, both of which include configurable reasoning budgets. Meta has also released the Llama 4 family of open-weight models, though those do not include a dedicated reasoning architecture equivalent to o3.

OpenAI, headquartered in San Francisco, operates ChatGPT, which the company has reported has more than 500 million weekly active users as of early 2025. The company has raised capital at a valuation of $300 billion following a funding round closed in March 2025.

What Happens Next

OpenAI has indicated that a full system card and technical documentation for o3 and o4-mini will be published following the initial release period, and the company is expected to present additional benchmark details at upcoming developer events.