OpenAI: GPT-5 Mini (batch)

GPT-5 Mini is a text-focused model from OpenAI, released on 7 August 2025 as a compact version of GPT-5. It is designed for lighter-weight reasoning tasks while retaining the instruction-following behaviour and safety tuning of the larger GPT-5 model. The batch variant covered here is served under the model id openai/gpt-5-mini:batch, so it targets workloads that tolerate batch style processing rather than strictly interactive traffic.

On the input side, the model accepts text, images and files, and it produces text as its only output modality. It supports a context window of 400,000 tokens, with a maximum output of 128,000 tokens, which is generous for long documents and multi-turn work. Its knowledge cutoff is 31 May 2024, so anything after that date has to be supplied in the prompt or through tools. Those tools are part of the picture: the model supports tool calling and structured outputs, so it can return JSON shaped to a schema and invoke functions you define.

The combination of a long context, image and file inputs, and structured output points at a few practical workloads. Document analysis is one: you can feed long reports, contracts or scanned material and ask for summarised or extracted fields returned in a fixed schema. Reasoning over mixed content is another, where a prompt pairs text with charts or screenshots and asks for a conclusion. Because tool calling is supported, the model also fits agent-style pipelines that need to fetch data, call internal APIs or run multi-step workflows, and lighter reasoning jobs where the full GPT-5 model would be more than the task requires.

You call it through api14's OpenAI-compatible API using the model id "openai/gpt-5-mini:batch". Requests follow the familiar chat completions shape, so existing OpenAI SDKs and clients work with a change of base URL and model id. The supported parameters are include_reasoning, max_tokens, reasoning, reasoning_effort, response_format, seed, structured_outputs, tool_choice, tools and verbosity. Together these let you cap output length, control how much reasoning effort the model spends, request reproducible sampling with a seed, enforce a response format or structured output schema, define and select tools, surface reasoning content, and tune verbosity.

The main thing to weigh is the batch nature implied by the model id, which suits asynchronous or throughput-oriented jobs better than tightly interactive ones, though the same API surface is used either way. The knowledge cutoff of 31 May 2024 means recent facts must come from your prompt or tools rather than the model's own training. If a task needs capabilities beyond text output, or heavier reasoning than this compact model is intended for, a different model in the catalogue may be the better fit.