gpt-4o-mini-tts-2025-03-20

Overview. gpt-4o-mini-tts-2025-03-20 is OpenAI’s March 20, 2025 snapshot of a text-to-speech model built on GPT-4o Mini. It takes text input and returns audio on the speech endpoint, not chat or realtime APIs. Developers reach for it when they need compact prompt-to-speech synthesis rather than a general-purpose language model.

Key strengths. The model is purpose-built for speech synthesis: text in, audio out, with streaming playback suited to on-demand spoken responses. Documented uses include narration, multilingual spoken audio, and realtime audio output. Optional audio streaming lets callers start playback without waiting for a full render on the speech API.

When to use it. Use this snapshot for simple text-to-speech jobs when you need spoken output of short prompts and do not need chat, tools, or multi-step reasoning. It fits interactive product flows—notifications, read-aloud helpers, lightweight voice UX—where input stays within a few thousand tokens. Prefer it only while the snapshot remains available and migration off deprecated speech APIs is still staged.

Limitations. Input is capped at 2,000 tokens, far below typical chat context windows, and output is audio only with no text completion, tool calling, or structured output path. The speech endpoint does accept an instructions string for steering voice style, but not a Chat Completions system message. OpenAI has deprecated this snapshot for API shutdown on January 6, 2027, pointing callers toward gpt-realtime-2.1-mini. It is not a Chat Completions, Responses, or Realtime model.

Sovereignty note. OpenAI is a U.S. creator, so hosted inference carries CLOUD Act exposure and is not Sovereign. Closed proprietary weights, a U.S. default endpoint, and only an eligible-customer option to process API data in Europe make this a poor fit for regulated workloads that require guaranteed EU data residency or sensitivity-safe routing.