OpenAI: o3

OpenAI o3 is a text-focused model from OpenAI, released by the provider on 16 April 2025. It sits in the text category and is described by OpenAI as a well-rounded and powerful model across domains, with particular strength in math, science, coding and visual reasoning, as well as technical writing and instruction-following. Its knowledge cutoff is 30 June 2024, so anything after that date is outside what the model learned during training.

On the input side, o3 accepts text, images and files, which means you can send a prompt that combines written instructions with a screenshot, diagram, chart or document. Output is text only. The context window is 200,000 tokens, and the maximum output is 100,000 tokens, so long inputs and long generations are both possible within the same request. The combination of image and file input with a large context makes it suitable for tasks where the model needs to read source material before answering rather than relying on memory alone.

Practical uses follow from those capabilities. Coding tasks, including reading existing code and reasoning about changes, are a natural fit, as are math and science problems that need step-by-step working. Visual reasoning tasks such as interpreting charts, diagrams or screenshots work because images are accepted as input. Technical writing and instruction-following make it a reasonable choice for documentation, structured reports and any workload where the output must follow a precise format. Because files can be supplied directly, summarising or extracting from longer documents is also viable without pre-processing them into plain text yourself.

You call the model through api14's OpenAI-compatible API using the model id "openai/o3". The supported request parameters are include_reasoning, max_tokens, reasoning, response_format, seed, structured_outputs, tool_choice and tools. In practice that means you can control the output length with max_tokens, ask for reasoning to be surfaced with include_reasoning and reasoning, constrain the response shape with response_format or structured_outputs, make runs more reproducible with seed, and give the model functions to call through tools and tool_choice.

The main constraint to keep in mind is the knowledge cutoff of 30 June 2024, so questions about more recent events need to be handled by supplying the relevant material in the prompt, which the file and image inputs make straightforward. Output is text only, so if you need generated images or audio, a different model is required. For workloads that do not need reasoning, image input or a 200,000 token context, a smaller text model may be simpler and faster to work with.