OpenAI o3

OpenAI o3 is a reasoning model in the o-series, released on April 16, 2025 alongside o4-mini. It was the first o-series model with full agentic access to every tool inside ChatGPT — web search, Python interpreter, image generation, and file analysis — and it was trained via reinforcement learning to decide when and how to use them. The model also introduced "thinking with images": images become part of the chain of thought and can be manipulated (rotated, zoomed) during reasoning. In the API, o3 has a 200,000-token context window, 100,000-token max output, and a June 1, 2024 knowledge cutoff. The API identifier is o3 (snapshot o3-2025-04-16). Pricing: USD 2 per 1M input tokens (USD 0.50 cached) and USD 8 per 1M output tokens. The model has been succeeded by GPT-5 but remains available via the API. An OpenAI o3-pro variant was also released in June 2025.