OpenAI Makes o3 and o4-mini Reasoning Models Publicly Available

OpenAI released its o3 and o4-mini reasoning models to the public this week, giving developers and paying subscribers access to the company's most capable reasoning systems to date. The release marks the widest distribution of OpenAI's reasoning-focused model line since the original o1 model launched in late 2024.

What Happened

OpenAI made o3 and o4-mini available through its API and via ChatGPT for users on paid subscription tiers. The o3 model is positioned as the company's most powerful reasoning model, while o4-mini is designed to deliver strong reasoning performance at lower computational cost and faster response times. Both models are built to handle complex, multi-step problem-solving tasks across mathematics, coding, and scientific reasoning.

The o4-mini model also introduces a notable capability: it can use tools, including web search and code execution, during its reasoning process. This allows the model to retrieve information and run code as intermediate steps before producing a final answer, rather than relying solely on its training data.

Background

OpenAI first introduced its reasoning-focused model series with o1 in September 2024, which was followed by o1-mini and o3-mini in the months that followed. These models differ from standard large language models in that they are trained to spend additional compute time working through problems before responding, a technique the company refers to as "thinking."

The o3 model had previously been available only to a limited set of researchers and safety testers following its announcement in December 2024. At the time of its announcement, OpenAI reported that o3 achieved high scores on several established benchmarks, including the ARC-AGI benchmark, which tests a model's ability to solve novel visual reasoning problems. OpenAI reported o3 scored approximately 87.5 percent on the standard ARC-AGI evaluation, compared to a human average of approximately 85 percent on the same set.

OpenAI, founded in 2015 and headquartered in San Francisco, operates ChatGPT, which the company has reported surpassed 400 million weekly active users as of early 2025. The company competes in the reasoning model segment with Google DeepMind, whose Gemini 2.0 series includes reasoning-capable variants, and with Anthropic, whose Claude 3.7 Sonnet model includes an extended thinking mode.

What It Means in Practice

Developers accessing o3 and o4-mini through the API can integrate the models into applications requiring complex reasoning, including automated coding assistants, research tools, and data analysis pipelines. The o4-mini model's tool-use capability during reasoning steps expands the range of tasks it can perform without requiring developers to build separate orchestration layers.

OpenAI has structured pricing to reflect the difference in capability and compute requirements between the two models. O3 carries a higher per-token cost, while o4-mini is priced to be more accessible for high-volume applications. Specific per-token pricing figures were published by OpenAI on its API pricing page at the time of release.

Access to o3 within ChatGPT is currently limited to Plus, Pro, and Team subscribers. Enterprise and Education account holders also receive access under their existing agreements. Free-tier users do not have access to o3 at launch, though o4-mini is available to a broader set of users given its lower resource requirements.

What Comes Next

OpenAI has indicated it plans to continue updating its reasoning model line, with future iterations expected to further integrate tool use and multimodal capabilities, including image input, which o4-mini already supports at launch.