Openai announced the release of the O3 and O4-Mini on Wednesday. This is a new AI inference model designed to pause and work with questions before responding.
The company calls O3 an unprecedented cutting edge inference model, surpassing the company’s previous models in tests that measure mathematics, coding, inference, science and visual understanding abilities. Meanwhile, the O4-MINI offers what Openai is a competitive trade-off between price, speed and performance. There are three factors that developers consider when choosing an AI model to enhance their applications:
Unlike previous inference models, O3 and O4-MINI can generate responses using ChatGPT’s tools such as web browsing, Python code execution, image processing, and image generation. Starting today, the O4-mini variant called “O4-mini-high” that spends more time creating answers to improve reliability is available to Sopenai Pro and Team Plan subscribers.
The new model is part of Openai’s efforts to beat Google, Meta, Xai, Anthropic and Deepseek in the Cutthroat Global AI race. Openai was the first to release the O1 AI inference model, but its competitors quickly matched or exceeded the performance of Openai’s lineup. In fact, inference models are beginning to dominate the field as AI Labs aims to deliver more performance from the system.
O3 has hardly been released on ChatGpt. Openai CEO Sam Altman signaled in February that the company intended to devote more resources to sophisticated alternatives that incorporate O3 technology. However, the competitive pressure seemed to have driven Openai to ultimately turn the course back.
According to Openai, the O3 achieves cutting-edge performance in SWE Bench Verification (no custom scaffolding), a test that measures coding capabilities, earning 69.1%. The O4-MINI model achieves similar performance, earning 68.1%. Openai’s next best model, the O3-Mini, scored 49.3% in Tests, while Claude 3.7 Sonnet scored 62.3%.
Openai claims that O3 and O4-Mini are the first models that can be “thinked in images.” In reality, users can upload images to ChatGPT, such as whiteboard sketches and diagrams from PDFs. The model analyzes the images during the “conceived chain” phase before responding. Thanks to this new feature, O3 and O4-MINI can understand blurry and low-quality images and perform tasks such as zooming and rotating images.
Beyond image processing capabilities, O3 and O4-MINI can run and run Python code directly in the browser via ChatGPT’s Canvas functionality and search the web when asked about current events.
In addition to CHATGPT, all three models, O3, O4-MINI and O4-MINI-HIGH, will be available via OpenAI’s developer endpoints, chat completion APIs, and response APIs, allowing engineers to build applications with company models at usage-based rates.
Openai charges developers a relatively low price for O3, giving its performance improvements, with an input token per million (approximately 750,000 words, longer than the Rods of the Rings series) and an output token per million of $40. For O4-MINI, Openai charges the same bill as the O3-Mini, $1.10 per million and $4.40 per million output token.
In the coming weeks, Openai plans to release O3-Pro, a version of O3 that uses more computing resources to create answers, exclusively for ChatGPT Pro subscribers.
Openai CEO Sam Altman shows that O3 and O4-Mini could be the last standalone AI inference model for CHATGPT before GPT-5.
Source link
