OpenAI: o3 Mini (batch)

OpenAI o3 Mini (batch) is a text model from OpenAI, released by the provider on 31 January 2025. It is a reasoning focused model, cost efficient and optimized for STEM tasks, with particular strength in science, mathematics and coding. The batch label indicates how the model is served through api14 rather than a separate capability tier, so you can expect the same core behaviour as o3 Mini while submitting work in batch.

The model accepts text and file inputs and returns text only. It has a context window of 200,000 tokens and a maximum output of 100,000 tokens, which leaves considerable room for long inputs and long generated answers in the same request. Its knowledge cutoff is 31 October 2023, so anything after that date has to come from your prompt or from tools you attach.

Because it is tuned for STEM reasoning, natural fits include mathematical problem solving, step by step scientific analysis and code generation or debugging. File input makes it useful for working over documents, for example extracting structured data from a specification or answering questions about a paper. The reasoning parameter lets you trade depth of internal reasoning against latency, and include_reasoning exposes that reasoning in the response when you want to inspect it. Tool calling extends this to retrieval, calculators or code execution, and structured outputs plus response_format help when a downstream system needs predictable JSON shaped results.

You call it through api14's OpenAI-compatible API using the model id "openai/o3-mini:batch". The supported request parameters are include_reasoning, max_tokens, reasoning, response_format, seed, structured_outputs, tool_choice and tools. Setting a seed gives you more reproducible sampling, max_tokens caps the generated output, and tools with tool_choice control whether and how the model calls functions you define. No other parameters should be assumed to work with this model.

One constraint worth noting is the knowledge cutoff of 31 October 2023, so tasks that depend on recent events need fresh context supplied in the request. If your workload is purely conversational or does not involve reasoning heavy steps, a general purpose text model may be simpler and faster. If you do need strong STEM reasoning over large inputs with tool use, o3 Mini (batch) is a reasonable default to evaluate first.