OpenAI · 2026-07-06 · major
OpenAI gpt-realtime-2.1 — voice model gains reasoning-effort dial and a mini variant
OpenAI shipped gpt-realtime-2.1 and gpt-realtime-2.1-mini on the Realtime API with a configurable reasoning-effort dial and better alphanumeric, silence, and interruption handling. Text input is $4 per 1M tokens; audio input is $32 per 1M.
gpt-realtime-2.1 adds a reasoning-effort dial and better silence/interruption handling to OpenAI's Realtime API.
Quick facts
| Maker | OpenAI |
|---|---|
| Models | gpt-realtime-2.1 and gpt-realtime-2.1-mini |
| Context window | 128K tokens (32K max output) |
| Text price | $4 in / $0.40 cached / $24 out per 1M |
| Audio price | $32 in / $0.40 cached / $64 out per 1M |
| What's new | Configurable reasoning effort, better silence and interruption handling |
| Availability | Realtime API (function calling supported) |
Pricing
| Text input | $4.00 / 1M tokens |
|---|---|
| Text cached input | $0.40 / 1M tokens |
| Text output | $24.00 / 1M tokens |
| Audio input | $32.00 / 1M tokens |
| Audio cached input | $0.40 / 1M tokens |
| Audio output | $64.00 / 1M tokens |
What is it?
gpt-realtime-2.1 is OpenAI's updated Realtime voice model, joined by a distilled gpt-realtime-2.1-mini for latency-sensitive agents. Both improve on gpt-realtime-2 for alphanumeric recognition, silence and noise handling, and interruption behavior, and both expose a configurable reasoning-effort setting per call.
How does it work?
The Realtime API streams audio in and out over a persistent connection, and gpt-realtime-2.1 lets developers dial reasoning-effort so a single agent can 'think longer' on hard turns and answer instantly on easy ones. Function calling is supported so the model can call tools mid-conversation, and the 128K context window with 32K max output keeps long back-and-forth sessions in one call.
Why does it matter?
Voice-agent builders finally get a per-turn reasoning knob without leaving the Realtime API — no more juggling a fast model and a smart model. The alphanumeric fix targets the most common failure mode in production voice agents (misheard phone numbers, order IDs, confirmation codes), and audio pricing is unchanged so migrating from gpt-realtime-2 is a model-name swap.
Who is it for?
Voice-agent developers, telephony and IVR teams, support automation.
Frequently asked questions
- What did gpt-realtime-2.1 change from gpt-realtime-2?
- gpt-realtime-2.1 improves alphanumeric recognition (phone numbers, codes, order IDs), silence and background-noise handling, and how the model handles being interrupted mid-response. It also adds a configurable reasoning-effort dial so voice agents can trade latency for deeper reasoning per call.
- How much does gpt-realtime-2.1 cost?
- gpt-realtime-2.1 charges $4 per 1M input text tokens, $0.40 cached, and $24 per 1M output text tokens. Audio tokens are $32 in, $0.40 cached, and $64 out per 1M — the same audio-token pricing as prior Realtime models, so most existing voice-agent projects can swap models without a pricing rewrite.
- What is gpt-realtime-2.1-mini for?
- gpt-realtime-2.1-mini is a faster, cheaper distilled variant of gpt-realtime-2.1 aimed at low-latency voice agents — think phone IVR, drive-thru, and lightweight assistants. It keeps the same Realtime API interface and function-calling support so a project can start on the full model and fall back to mini per call.
- Where do developers get gpt-realtime-2.1?
- gpt-realtime-2.1 and its mini variant are available today on the OpenAI Realtime API. Both models support function calling and cached input pricing, and OpenAI's changelog and model docs list gpt-realtime-2.1 and gpt-realtime-2.1-mini as the recommended targets for new voice-agent projects.
Try it
Set model: 'gpt-realtime-2.1' on the OpenAI Realtime API — see developers.openai.com/api/docs/models/gpt-realtime-2.1.