gpt-realtime-mini, gpt-audio-mini and gpt-image-1-mini

2025-10-06 · Publisher : OpenAI · Domain : Artificial Intelligence

At DevDay, OpenAI opens three more economical models in the API: gpt-realtime-mini and gpt-audio-mini for speech-to-speech, and gpt-image-1-mini for image generation.

What changes

  • gpt-realtime-mini: lightweight version of gpt-realtime, about 70% cheaper with comparable quality, via WebRTC, WebSocket and SIP.
  • gpt-audio-mini: economical version of gpt-audio for audio input and output in Chat Completions.
  • gpt-image-1-mini: about 80% cheaper than gpt-image-1 for a similar output, for image generation and editing.
  • The same day: GPT-5 pro and Sora 2 with Sora 2 Pro become available in the API (see the GPT-5 and Sora 2 entries).
  • New December 2025 snapshots (2025-12-15) improve the reliability and voice fidelity of the mini audio models.

### Specifications - gpt-realtime-mini: context 32,000 tokens, output 4,096 tokens, text, image and audio inputs, text and audio outputs, cutoff October 1, 2023. - gpt-audio-mini: context 128,000 tokens, output 16,384 tokens, cutoff October 1, 2023. - gpt-image-1-mini: text and image inputs, image output. ### Pricing Price per million tokens (USD), according to the current documentation. | Model | Text input | Text output | Other | | --- | --- | --- | --- | | gpt-realtime-mini | 0.60 (cache 0.06) | 2.40 | audio prices not included | | gpt-audio-mini | 0.60 | 2.40 | audio prices not included | | gpt-image-1-mini | 2 | not applicable | image: 2.50 input, 8 output | Per 1024x1024 image: about 0.005 (low), 0.011 (medium), 0.036 (high) dollars. ### Benchmarks No score published by the vendor for these models in the sources consulted. ### Key takeaways These mini variants lower the cost of voice agents and image pipelines at volume, at the price of lower quality than the full models.

Links

Tags : Audio, Image, OpenAI, Temps réel, DevDay, API