OpenAI Releases o3 and o4-mini Reasoning Models Publicly

OpenAI released two new reasoning-focused AI models, o3 and o4-mini, making them available to the public through its API and ChatGPT platform. The release marks the first time OpenAI has offered models from its o-series reasoning line to general users outside of limited research previews.

What Happened

OpenAI confirmed the public availability of o3 and o4-mini on April 16, 2025. Both models are accessible via the OpenAI API for developers and through ChatGPT for subscribers on Plus, Pro, and Team plans. The o3 model represents OpenAI's most capable reasoning system to date in the o-series line, while o4-mini is positioned as a faster, more cost-efficient alternative designed for high-volume applications.

The company stated that o4-mini delivers reasoning performance competitive with larger models at a fraction of the computational cost. OpenAI did not disclose specific pricing tiers at the time of the announcement beyond confirming API access would follow its standard usage-based billing structure.

Background

OpenAI introduced its o-series reasoning models in September 2024 with the release of o1, a model specifically designed to spend more time processing complex problems before generating a response. The approach, which OpenAI describes as "thinking before answering," was developed to improve performance on tasks requiring multi-step logic, mathematics, coding, and scientific reasoning.

O1 was followed by o1-mini, a smaller variant, and o3 was previewed in December 2024 during a series of live-streamed announcements. At the time of the December preview, o3 achieved a score of 87.5 percent on the ARC-AGI benchmark, a test designed to measure general reasoning capabilities. That result drew significant attention within the research community, as prior models had scored substantially lower on the same benchmark.

O4-mini had not been publicly previewed before the April 2025 release. OpenAI described it as incorporating tool use capabilities, including the ability to use web search, run code, and analyze images within a single reasoning chain.

Key Capabilities

According to OpenAI's technical documentation released alongside the models, both o3 and o4-mini support multimodal inputs, meaning they can process text and images together. O4-mini is noted for its ability to use tools autonomously during its reasoning process, selecting when to search the web or execute code without explicit prompting from the user.

OpenAI stated that o3 outperforms its previous flagship reasoning model, o1, on a range of benchmarks including competition mathematics, PhD-level science questions, and software engineering tasks. The company also reported that o4-mini matches or exceeds o1's performance on several of those same benchmarks while operating at lower latency and cost.

Safety evaluations conducted by OpenAI prior to release indicated that o3 required additional mitigations in areas related to biological and chemical information compared to earlier models, according to the company's published safety card. OpenAI said those mitigations were applied before the public release.

What It Means in Practice

Developers accessing o3 through the API can deploy the model in applications requiring complex, multi-step reasoning tasks. O4-mini's lower cost structure makes it suitable for applications where reasoning is needed at scale, such as automated code review, document analysis, or customer-facing query systems that require logical inference.

Enterprise customers already using OpenAI's API under existing agreements will have access to both models under current contract terms, according to OpenAI's release documentation. Usage limits for ChatGPT subscribers vary by plan tier.

OpenAI has not announced a release date for o3-pro, a higher-capability variant of o3 referenced in the company's model roadmap, though the company indicated it is in preparation for release in the coming weeks.