OpenAI: GPT-4o (2024-05-13)

GPT-4o (2024-05-13) is a text model from OpenAI. The "o" stands for "omni", and this release is dated 13 May 2024. It is built to accept both text and image inputs while returning text outputs, and OpenAI describes it as maintaining the intelligence level of GPT-4 Turbo while being faster. If you are choosing a general purpose OpenAI model for a new integration, this is the version identified by the model id openai/gpt-4o-2024-05-13.

On inputs, the model accepts text, images and files, and it produces text only. The context window is 128,000 tokens, so long documents, large code files or extended conversations can be supplied in a single request. The maximum output is 4,096 tokens, which matters when you plan long generations, since the reply will be truncated at that point regardless of how much room remains in the context. Its knowledge cutoff is 31 October 2023, so anything after that date has to come from the prompt, from tools or from a web search rather than from the model's own training.

Those capabilities point to a few practical uses. Because it reads images as well as text, it suits tasks such as extracting structured data from screenshots, scanned pages or diagrams, describing visual content, or answering questions about a chart alongside a written brief. The large context window makes it workable for document question answering, summarising long reports, and reviewing sizable codebases in one pass. Support for tools and structured outputs makes it a reasonable choice for agent style workflows and for pipelines where the model must return JSON that another service parses.

Through api14 you call it with the model id openai/gpt-4o-2024-05-13 using the OpenAI-compatible API, so existing OpenAI SDKs and clients work with a change of base URL and key. The supported request parameters are frequency_penalty, logit_bias, logprobs, max_completion_tokens, max_tokens, prediction, presence_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_logprobs, top_p and web_search_options. Tools and tool_choice cover function calling, response_format and structured_outputs cover constrained output shapes, and seed supports more reproducible sampling.

Two limits are worth planning around. The 4,096 token output ceiling is modest for long form writing, so tasks that need very lengthy single responses may suit another model better, and the October 2023 knowledge cutoff means time sensitive questions should be handled with tools or web search rather than relying on the model's memory.