Chat with OpenAI o3

OpenAI o3 is currently the most powerful reasoning model, pushing the frontier across multiple domains including coding, mathematics, science, and visual perception. Not only has it made significant breakthroughs in multi-domain capabilities, but it has also set new records on several authoritative benchmarks (such as achieving new SOTA on Codeforces, SWE-bench, and MMMU, and in expert evaluations, making 20% fewer major errors than o1 on complex, real-world tasks). It is especially adept at handling problems that require complex analysis and deep reasoning, and excels at visual analysis tasks.

OpenAI o3 is the first model in its series that can truly "think with images." Users can upload images to ChatGPT, such as whiteboard sketches or diagrams from PDFs, and o3 will analyze these images during its "chain-of-thought" phase before answering. This revolutionary capability enables o3 to understand blurry and low-quality images and perform tasks such as zooming or rotating images as it reasons, dramatically expanding its problem-solving toolkit.

OpenAI o3 underwent a comprehensive safety upgrade, including a complete rebuild of safety training data and the addition of new refusal prompts in areas such as biological threats (biorisk), malware generation, and jailbreaks. We have also developed system-level mitigations to flag dangerous prompts in frontier risk areas. Internal evaluations show that OpenAI o3 achieves strong performance on refusal benchmarks and dangerous prompt detection, and remains below the Framework’s “High” threshold across all major risk categories.