OpenAI has launched its latest models, o3 and o4-mini, described as the company’s smartest and most capable releases to date. The new models are now available to ChatGPT users with full access to tools like web browsing, file analysis with Python, visual input reasoning, and image generation.
OpenAI says o3 is its most powerful reasoning model yet, delivering state-of-the-art performance in coding, math, science, and visual tasks. The model significantly reduces major errors compared to its predecessor, OpenAI o1, and shows notable improvements in programming, business consulting, and creative ideation.
Meanwhile, o4-mini offers fast, cost-efficient reasoning. Optimized for high-volume use, o4-mini performs strongly in math, coding, and visual tasks, outperforming its predecessor, o3-mini, in benchmarks like AIME 2025 and data science evaluations. Both new models also demonstrate improved instruction-following capabilities and more personalized responses by referencing past conversations.
For the first time, OpenAI’s models can think with images, integrating visual content into their reasoning processes. The models can analyze photos, charts, and low-quality visuals, using tools like rotation and zooming to assist problem-solving.
OpenAI trained both models to not only use tools but also reason about when and how to deploy them. This allows the models to complete complex tasks independently, such as conducting web searches, writing code, generating graphs, and delivering full analyses—all within about a minute.
Safety improvements accompanied these advances. OpenAI rebuilt its safety training datasets and deployed a reasoning LLM monitor to better detect dangerous prompts, reporting strong performance on internal refusal benchmarks. Evaluations confirmed that both o3 and o4-mini remain below OpenAI’s “High” risk thresholds across biological, cybersecurity, and self-improvement categories.
Both models are now available to ChatGPT Plus, Pro, and Team users, replacing previous versions. Enterprise and Edu users will gain access within a week. Developers can access o3 and o4-mini through OpenAI’s Chat Completions API and the newly introduced Responses API.
OpenAI also launched Codex CLI, a lightweight, open-source coding agent designed for direct terminal use, and announced a $1 million grant program to support Codex CLI projects, offering grants in $25,000 increments via API credits.
OpenAI says its upcoming plans involve merging the strengths of the o-series and GPT-series models to create tools that combine natural conversation with advanced, proactive reasoning.
Image: OpenAI
This article, “OpenAI Introduces o3 and o4-mini Models with Enhanced Reasoning and Full Tool Access” was first published on Small Business Trends