OpenAI unveils advanced o3 and o4-mini AI models with autonomous tool usage

OpenAI has unveiled its latest reasoning models, o3 and o4-mini, marking a significant leap in AI capabilities. These models are the first to independently utilize all ChatGPT tools, including web browsing, Python, image understanding, and image generation, to solve complex, multi-step problems.

Key Features of o3 and o4-mini

• Visual Understanding: The models can "think with images," processing visual information to enhance reasoning. They can even interpret low-quality or blurry images.
• Autonomous Tool Usage: Unlike previous models, o3 and o4-mini can agentically employ tools like web browsing and Python without user intervention, paving the way for more autonomous AI assistants.
• Improved Performance: Benchmarks show these models outperform their predecessors in coding, math, and science tasks, even without additional tools.

Demo Highlights

During a livestream, OpenAI demonstrated o3 analyzing a scientific research poster. The model browsed the web and zoomed into image details to draw conclusions not explicitly stated in the poster, showcasing its advanced reasoning and tool integration.

Availability

The models are now available to ChatGPT Plus, Pro, and Team subscribers, replacing older versions in the model picker. Developers can also access them via the API. Pro users will gain access to o3-pro in the coming weeks.

Safety and Transparency

OpenAI emphasized rigorous safety testing under its Preparedness Framework. Detailed evaluations are available in the system card.

Bonus: Codex CLI and Grants

OpenAI also launched Codex CLI, an open-source coding agent for local terminal use, supporting o3 and o4-mini (with GPT-4.1 support coming soon). It’s available on GitHub. Additionally, a $1 million initiative offers grants for early projects, with proposals accepted via OpenAI’s form.

For more AI updates, subscribe to Innovation, OpenAI’s weekly newsletter.