Cost-efficient version of GPT Audio.
GPT Audio Mini (gpt-audio-mini) is OpenAI's cost-efficient audio-capable model for the Chat Completions API, launched in January 2026 alongside the full gpt-audio. It shares the same upgraded decoder, multi-turn audio conversation, and base64-encoded audio input/output capabilities as its larger sibling, but at approximately 75% lower pricing per token, making it a practical choice for high-volume voice applications where per-call cost matters.
Like gpt-audio, gpt-audio-mini supports mixing text and audio modalities within a single HTTP Chat Completions call. Developers can send audio (WAV/MP3 via base64) alongside text and receive either or both modalities in the response. The mini variant shares the same 128K context window, tool/function calling support, and streaming capabilities — the difference is purely in model scale, reflected in lower output quality but significantly reduced cost.
| Type | Rate |
|---|---|
| Input | $0.648 / 1M tokens |
| Output | $2.59 / 1M tokens |
| Audio input | $10.80 / 1M tokens |
| Audio output | $21.60 / 1M tokens |
| Release date | January 19, 2026Design for Online / OpenRouter ↗ | verified |
| Context window | 128,000 tokensOpenRouter ↗ | verified |
| Max output tokens | 16,384 tokensOpenRouter ↗ | verified |
| Supported modalities | Input: text, audio; Output: text, audiodevelopers.openai.com ↗ | verified |
| Supported capabilities | Tools, function calling, response_format, streaming, audioplatform.openai.com ↗ | verified |
| License | Proprietary (OpenAI API Terms) | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
Audio quality (upgraded decoder) gpt-audio-mini uses the same upgraded decoder as gpt-audio for more natural-sounding voices with better voice consistency, but compresses model scale for cost efficiency. No published benchmark scores are available for the mini variant specifically. | Upgraded decoder — same gpt-audio base with reduced scale | Design for Online ↗ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "openai/gpt-audio-mini",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'Send base64-encoded WAV/MP3 audio in the message content. The model replies with text and/or audio — identical API pattern to gpt-audio, just at a lower price tier.
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "openai/gpt-audio-mini",
"modalities": ["text", "audio"],
"audio": { "voice": "alloy", "format": "wav" },
"messages": [
{
"role": "user",
"content": [
{ "type": "text", "text": "What does this audio say?" },
{ "type": "input_audio", "input_audio": { "data": "<BASE64_AUDIO>", "format": "wav" } }
]
}
]
}'OpenAI released two audio-capable Chat Completions models: gpt-audio (full) and gpt-audio-mini (cost-efficient). Both feature an upgraded decoder for more natural voices compared to the earlier gpt-4o-audio-preview. The mini variant is priced ~75% lower per token, targeting high-volume voice applications.
gpt-audio-mini is a smaller, more cost-efficient variant of gpt-audio. Both share the same upgraded decoder, 128K context window, tool calling, and audio input/output modalities. The mini model produces lower-quality audio output but costs approximately 75% less per token ($0.15/M input text + $10/M audio vs $2.50/M input text + $100/M audio for gpt-audio).
No. gpt-audio-mini uses the HTTP-based Chat Completions API, which is designed for request-response patterns. For sub-500ms real-time speech-to-speech, use gpt-realtime-mini via OpenAI's Realtime API / WebRTC instead.
Yes. gpt-audio-mini fully supports tool and function calling, including audio in messages alongside tool definitions. You can send audio (e.g. a recording with a booking reference) and instruct the model to extract structured data via function calls — the same way you would with a text-only model.
Audio input supports WAV and MP3 formats encoded as base64 in the input_audio content part. Audio output supports WAV format with configurable voice options (alloy, nova, etc.) set via the audio parameter.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | Design for Online / OpenRouter ↗ | 2026-05-29 | verified |
| Context window | OpenRouter ↗ | 2026-05-29 | verified |
| Max output tokens | OpenRouter ↗ | 2026-05-29 | verified |
| Supported modalities | developers.openai.com ↗ | 2026-05-29 | verified |
| Supported capabilities | platform.openai.com ↗ | 2026-05-29 | verified |
| License | — | — | verified |
| Audio quality (upgraded decoder) | Design for Online ↗ | 2026-05-29 | to verify |
| OpenAI launches gpt-audio and gpt-audio-mini | Design for Online / OpenRouter ↗ | 2026-05-29 | verified |