GPT-Realtime mini
GPT-Realtime mini is a cost-efficient realtime model that responds to audio and text inputs in realtime, built for voice features that need low latency and workable unit economics at high volume.
import { gateway } from '@ai-sdk/gateway';
export async function POST() { const { token, url } = await gateway.experimental_realtime.getToken({ model: 'openai/gpt-realtime-mini', });
return Response.json({ token, url, tools: [] });}Playground
Try out GPT-Realtime mini by OpenAI. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.
GPT-Realtime miniVoice agent
Talk to a realtime agent. It listens to your voice and replies with audio.
Start the session and ask the agent something.
Providers
Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.
More models by OpenAI
| Model |
|---|
| 1.1M | 3.6s | 90tps | $5/M | $30/M | Read: $0.5/M Write: $6.25/M | $10/K + input costs | +4 |  | 07/09/2026 | |||
| 1.1M | 0.9s | 161tps | $1/M | $6/M | Read: $0.1/M Write: $1.25/M | $10/K + input costs | +4 | | 07/09/2026 | |||
| 1M | 0.5s | 136tps | $5/M | $30/M | Read: $0.5/M Write: — | $10/K + input costs | +4 |  | 04/24/2026 | |||
| 400K | 1.0s | 139tps | $0.75/M | $4.50/M | Read:$0.07/M Write:— | $10/K + input costs | +4 | | 03/17/2026 | |||
| 1.1M | 3.0s | 91tps | $2.50/M | $15/M | Read: $0.25/M Write: — | $10/K + input costs | +4 | | 03/05/2026 | |||
| 400K | 3.9s | 161tps | $0.25/M | $2/M | Read:$0.03/M Write:— | $14/K + input costs | +4 | | 08/07/2025 |
About GPT-Realtime mini
GPT-Realtime mini launched on October 10, 2025 at OpenAI's DevDay as the cost-efficient entry in the gpt-realtime family. GPT-Realtime mini responds to audio and text inputs in realtime, covering interactive voice use cases that don't need a flagship model on every turn.
The economics are the point. Voice features strain budgets when every session invokes a top-tier model. GPT-Realtime mini exists so in-app assistants, kiosks, and voice helpers can run continuously without the flagship rate, while keeping the low-latency conversational feel that makes voice interfaces usable.
Through AI Gateway, GPT-Realtime mini connects over WebSocket using the AI SDK's realtime hook. Your server mints a short-lived token, the browser handles microphone capture and playback through the hook, and AI Gateway applies the same observability and spend controls as your other models.
What To Consider When Choosing a Provider
- Configuration: Realtime support on AI Gateway is in beta through AI SDK 7, connecting over WebSocket with a short-lived server-minted token. For agents that need heavier reasoning or complex tool chains, step up to
gpt-realtime-1.5orgpt-realtime-2; the mini variant is tuned for cost and speed. - Zero Data Retention: AI Gateway does not currently support Zero Data Retention for this model. See the documentation for models that support ZDR.
- Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.
When to Use GPT-Realtime mini
Best for
- High-volume voice features: Per-session cost decides viability and the mini rate keeps features running
- In-app voice assistants: Routine questions and commands handled with realtime responses
- Voice prototypes: Realtime experiments without committing to flagship pricing
- Always-on voice surfaces: Kiosks, devices, and embedded experiences running continuous interaction
Consider alternatives when
- Mission-critical voice agents:
gpt-realtime-1.5brings stronger instruction following and tool calling - Reasoning-heavy conversations:
gpt-realtime-2adds configurable reasoning effort for complex workflows - Single-direction audio jobs:
gpt-4o-mini-transcribeandtts-1handle transcription and speech generation directly
Conclusion
GPT-Realtime mini answers the cost question that stalls most voice features. Use the mini variant for high-volume, routine interactions through AI Gateway, and reserve gpt-realtime-1.5 or gpt-realtime-2 for the conversations that earn a flagship model.