OpenAI has released two new AI models — o3 and o4-mini — aimed at enhancing reasoning capabilities across applications. The launch introduces expanded features in ChatGPT and new options for developers looking for a balance between performance and cost.
The OpenAI o3 and o4-mini models are designed to handle complex reasoning by pausing to evaluate a question before answering, a technique commonly referred to as chain-of-thought prompting. According to OpenAI, the o3 model is the strongest performer to date in areas such as coding, math, visual understanding, and scientific problem solving. It outperforms earlier models and achieves top scores in evaluations like SWE-bench verified, a benchmark for coding ability.
o3 scored 69.1% on that test, while o4-mini followed closely with 68.1%, marking a significant leap from the 49.3% score posted by OpenAI’s earlier o3-mini. For comparison, competitor Claude 3.7 Sonnet scored 62.3%, showing that OpenAI’s new models have narrowed — or even overtaken — the competition in this area.
Beyond code and math, both models bring a new feature set to ChatGPT, including the ability to process images as part of their reasoning. Users can upload images such as diagrams or hand-drawn notes, and the AI will incorporate visual data into its analysis. These models can interpret low-resolution or blurry visuals and manipulate them — zooming, rotating, or examining specific elements — to extract meaning.
Developers can access o3 and o4-mini via the Chat Completions and Responses APIs, with usage-based pricing. Notably, the pricing for o3 is set at $10 per million input tokens and $40 for output tokens. In contrast, o4-mini is more budget-friendly, matching the rates of its predecessor at $1.10 per million input tokens and $4.40 for outputs. A third option, o4-mini-high, trades speed for more refined answers and is also available starting today.
All models are available to subscribers of ChatGPT’s Pro, Plus, and Team tiers. Within the ChatGPT interface, they can also access built-in tools like web search, Python execution, and image generation — making the o3 and o4-mini models particularly versatile for professional and technical use cases.
OpenAI says an o3-pro model is in the works, expected to deliver even more powerful responses by using additional compute resources. It will roll out soon to ChatGPT Pro users.
This release comes amid growing competition in the AI sector, with companies like Google, Meta, Anthropic, and xAI also pushing forward with advanced reasoning models. While OpenAI was among the first to launch such tools, the field has quickly evolved into a race for higher performance and broader capabilities.
These may be the final standalone reasoning models before OpenAI integrates them into its next-generation GPT-5 system — a model expected to unify traditional and reasoning-based AI into a single architecture.
