Key takeout
Openai’s O3 model can be thought of in images and solve complex, multi-step problems using tools such as web browsing and Python. O4-MINI is optimized for speed and cost-effectiveness, and outperforms previous models of mathematics, coding, and non-STEM tasks.
Please share this article
Openai today released two new AI models, the O3 and O4-Mini, expanding its inference and visual intelligence capabilities.
The O3 represents the company’s most advanced inference model to date, with O4-MINI improving performance in mathematics, coding and visual tasks at a lower cost.
The O3 is the first model in the O series that allows independent use of all available ChatGPT tools, including web browsing, Python, image understanding, generation, and more. Both models introduce “image-based thinking” and allow visual input to be integrated directly into the inference process.
The O3 model establishes new benchmarks in software engineering, mathematics and scientific reasoning, surpassing O1 in tasks that require detailed analysis, hypothesis generation, and interpretation of visual content. External testing shows that O3 reduces major errors by 20% compared to O1.
Optimized for high-throughput performance, O4-MINI ranks number one in benchmarks including AIME 2024 and 2025, showing strong accuracy across STEM and non-STEM fields.
Openai has also released the Codex CLI, a local coding agent for running models from terminals. The $1 million grant program supports developer buildings.
Both models were subjected to safety testing using OpenAI’s updated preparation framework, and assessments confirmed risk levels below the thresholds of biosecurity, cybersecurity, and self-improvement categories.
This model is currently available to ChatGpt Plus, Pro, and team users, replacing the O1 and O3-MINI. ChatGpt Enterprise and EDU customers will be able to access it next week. Free tier users can test O4-MINI using the “Think” option before the query.
Openai plans to launch the O3-Pro in the coming weeks, combining the features of O3 with full tool support for advanced inference tasks.
Please share this article
