OpenAI: GPT Audio Mini

GPT Audio Mini is an audio model from OpenAI, offered as a smaller, cost-efficient version of GPT Audio. It sits in the audio category and also covers text, so it is intended for workloads where spoken input and spoken output matter, not just typed conversation. The model id on api14 is openai/gpt-audio-mini, and the provider release date is 2026-01-19. Compared with the larger GPT Audio line, this snapshot uses an upgraded decoder for more natural sounding voices and maintains better voice consistency.

The model accepts text and audio as input and can produce text and audio as output. Its context window is 128,000 tokens, and the maximum output is 16,384 tokens. That combination means you can pass in a fairly long conversation or a lengthy audio-derived transcript alongside the audio itself, then receive either a written answer or a spoken one. Supported parameters include temperature, top_p, max_tokens, stop, seed, frequency_penalty, presence_penalty, logit_bias, logprobs and top_logprobs, so sampling and reproducibility controls are available, along with structured_outputs, response_format, tools and tool_choice for shaped responses and function calling.

Those capabilities point to a few practical patterns. Voice assistants and interactive agents are an obvious fit, since the model handles both sides of a spoken exchange and keeps voice consistency across turns. Transcription-style workflows that need a spoken reply rather than just text also work, as do accessibility features that read content aloud with a natural voice. Because tool calling and structured outputs are supported, you can build audio front ends that trigger backend functions, for example a phone agent that books appointments or looks up order status and answers verbally. Long context helps for multi-turn sessions or when the model needs a large transcript in front of it.

You call it through api14's OpenAI-compatible API using the model id openai/gpt-audio-mini. Point your existing OpenAI SDK or HTTP client at the api14 base URL, send a chat completion request with that id, and include audio content in the messages where needed. The request body accepts frequency_penalty, logit_bias, logprobs, max_tokens, presence_penalty, response_format, seed, stop, structured_outputs, temperature, tool_choice, tools, top_logprobs and top_p. Set response_format when you want a specific output shape, and use tools with tool_choice when the model should call functions. The page's live price table shows current rates.

One thing to weigh is output length: 16,384 tokens is the ceiling for a single response, so very long spoken answers or full audiobook-length generations are better split into segments. If your task is purely text, a text-only model may be simpler, and if you need the strongest audio quality OpenAI offers, the full GPT Audio model is the alternative. Otherwise, this mini variant covers the same input and output modalities at a smaller scale.