OpenAI: gpt-oss-120b (batch)

OpenAI: gpt-oss-120b (batch) is an open-weight language model from OpenAI. It is a Mixture-of-Experts (MoE) model with 117 billion total parameters, of which 5.1 billion are activated per forward pass. It was released by the provider on 2025-08-05 and sits in the text category. The "batch" suffix in the model id indicates the batch variant of this model as served on api14.

The model takes text in and returns text out, with a context window of 131,072 tokens and a maximum output of 117,964 tokens. Its knowledge cutoff is 2024-06-30. It is built for high-reasoning, agentic and general-purpose production use cases, and it supports reasoning controls, so you can decide how much reasoning effort the model applies and whether reasoning content is included in the response.

Those capabilities point to a fairly wide set of practical uses. The long context and large maximum output suit long document analysis, summarisation and generation tasks where the input and the answer are both large. The reasoning and agentic focus makes it a reasonable fit for multi-step problem solving, tool-driven workflows and structured extraction, since it supports tool calling and structured outputs. General-purpose text generation, such as drafting and transformation work, also follows from its text-in, text-out design.

You call it through api14's OpenAI-compatible API using the model id "openai/gpt-oss-120b:batch". Because the interface is OpenAI-compatible, existing OpenAI client libraries and request shapes work with it. The supported request parameters are frequency_penalty, include_reasoning, logit_bias, max_tokens, min_p, presence_penalty, reasoning, reasoning_effort, repetition_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_k and top_p. Sampling controls (temperature, top_p, top_k, min_p, the penalties, seed and stop), output limits (max_tokens), reasoning controls (reasoning, reasoning_effort, include_reasoning), and structured and tool use (response_format, structured_outputs, tools, tool_choice) are all available, so you can tune both the shape and the determinism of responses.

One thing to keep in mind is modality: input and output are text only, so if your task depends on images or audio, this model is not the right fit. The knowledge cutoff of 2024-06-30 also means anything after that date is outside its training data, and you should supply recent information in the prompt or via tools where it matters. Within those bounds, the combination of a large context window, a large maximum output and reasoning plus tool support covers most text production workloads.