Artificial intelligence continues to evolve at a rapid pace as tech companies push the boundaries of what's possible. In a significant advancement for AI reasoning capabilities, OpenAI has introduced two new models designed to think more thoroughly before responding to user queries.
New Reasoning Models Released
OpenAI has officially launched two new AI reasoning models: o3, which the company describes as its most powerful reasoning model to date, and o4-mini, a smaller and faster alternative that delivers impressive performance relative to its size and cost. These models follow OpenAI's recent introduction of its flagship GPT-4.1 model and represent the next evolution in the company's reasoning-focused AI systems. Unlike standard models, these reasoning models are specifically trained to think before they speak, taking additional processing time to deliver higher-quality responses.
New OpenAI Models Overview
| Model | Description | Key Features |
|---|---|---|
| o3 | Most powerful reasoning model | Visual reasoning, tool integration, enhanced performance |
| o4-mini | Smaller, faster, cost-effective | Visual reasoning, tool integration, optimized for speed |
| o4-mini-high | Higher performance variant | Enhanced capabilities while maintaining efficiency |
Tool Integration
- Web browsing
- Python coding
- Image understanding
- Image generation
Visual Understanding Integration
Perhaps the most significant advancement in these new models is their ability to think with images. Both o3 and o4-mini can now integrate visual information directly into their reasoning process, rather than simply describing what they see. This capability allows the models to work with sketches, whiteboards, and even low-quality or blurry images. The models can manipulate images as part of their analysis, including zooming in on specific elements or rotating images to better understand the content. This visual reasoning represents a substantial step forward in AI's ability to process and utilize information across different formats.
Agentic Tool Usage
Another major breakthrough is the models' ability to independently use all ChatGPT tools during their reasoning process. This includes web browsing, Python coding, image understanding, and image generation capabilities. OpenAI describes this as a step toward a more agentic ChatGPT that can independently execute tasks on your behalf. In practical terms, this means the models can autonomously decide which tools to employ to solve complex, multi-step problems—similar to how a person might reach for a calculator when needed. During a demonstration, researchers showed how o3 could analyze a scientific research poster, browse the internet for additional information, and zoom into different elements of the image to generate a comprehensive answer.
Performance Improvements
According to OpenAI, both o3 and o4-mini outperform their predecessors across various benchmarks, even without utilizing their new tool access capabilities. The models demonstrate enhanced instruction following and provide more useful, verifiable responses. While o3 represents the premium offering with maximum capabilities, o4-mini has been optimized to deliver strong performance at lower cost and higher speed, making advanced reasoning more accessible.
Availability and Access
The new models are available immediately to ChatGPT Plus, Pro, and Team subscribers, appearing in the model picker as o3, o4-mini, and o4-mini-high. They will replace the previous reasoning models o1, o3-mini, and o3-mini-high. Pro users will gain access to o3-pro in the coming weeks, but until then will continue to have access to o1-pro. Developers can also access these models through OpenAI's API.
Additional Announcements
Alongside these new models, OpenAI has launched Codex CLI, an open-source coding agent that runs locally in users' terminals. This tool provides a straightforward way to connect AI models, including o3 and o4-mini, to users' own code and tasks. Support for GPT-4.1 is expected soon. The company has also announced a $1 million initiative to support early projects using these technologies, offering grants in $25,000 increments via API credits.
Safety Considerations
OpenAI has emphasized that both new models underwent extensive safety testing before release. They were stress-tested under the company's safety program and evaluated according to its updated Preparedness Framework, with detailed results available in the complete system card.