OpenAI: GPT-4o-mini (batch)

GPT-4o-mini (batch) is a text generation model from OpenAI, offered here in its batch variant. It belongs to the GPT-4o family and is positioned as OpenAI's small model, sitting below the full GPT-4o. The provider released it on 18 July 2024, and its training knowledge runs up to 31 October 2023. If you have used other OpenAI models, the behaviour and interface will feel familiar.

The model accepts text, images and files as input, and it returns text only. Its context window is 128,000 tokens, which is enough to hold long documents, multi-turn conversations or a mix of prose and images in a single request. Output is capped at 16,384 tokens per response. Because image input is supported, you can send screenshots, diagrams or scanned pages alongside a text prompt and ask for a written answer. File input extends this to documents you want summarised, extracted from or questioned.

Those capabilities point to a fairly wide set of practical uses. Document question answering over long PDFs or reports, summarisation of meetings and threads, extraction of structured fields from unstructured text, classification and tagging, and drafting or rewriting content are all reasonable fits. Image input makes it usable for tasks like reading charts, transcribing visible text or describing what is in a screenshot. Tool calling and structured outputs mean it can sit inside an agent loop or feed a downstream system that expects a fixed schema, and the web search option lets a request pull in external context.

You call it through api14's OpenAI-compatible API using the model id "openai/gpt-4o-mini:batch". Because the surface is OpenAI-compatible, existing OpenAI SDKs and clients work with only a base URL and key change. The request parameters it supports are frequency_penalty, logit_bias, logprobs, max_tokens, prediction, presence_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_logprobs, top_p and web_search_options. That covers sampling controls, deterministic runs via seed, constrained decoding through response_format and structured_outputs, and tool use through tools and tool_choice.

Two things are worth checking before you commit. The knowledge cutoff is 31 October 2023, so anything more recent has to come from your prompt or from web_search_options. Output is text only, so if you need generated images or audio, this model is not the one. And since this is the batch variant, confirm that its latency profile matches your application; if you need interactive response times, compare against the standard non-batch entry for the same model.