GPT Realtime 2.1
A reasoning model with tool use that updates GPT-Realtime-2 with improved alphanumeric recognition, silence and noise handling, and interruption behavior for speech-to-speech voice-agent workflows.
text + audio → text + audio · made by OpenAI
$4→$24per Mtok in / out
OpenAI · 3 sellers
Sold by 3 ways
| Seller | Lane | Rate |
|---|---|---|
| Microsoft Azure AI cloud | standard | $4per Mtok in$24per Mtok out$0.4per Mtok cached$0.4per Mtok cached, audio$0.5per Mtok cached, image$32per Mtok in, audio$5per Mtok in, image$64per Mtok out, audioraw.githubusercontent.com · read 2026-09-16 |
| OpenAI api | standard | $4per Mtok in$24per Mtok out$0.4per Mtok cached, audio$32per Mtok in, audio$64per Mtok out, audiodevelopers.openai.com · read 2026-08-24 |
| Vercel AI Gateway aggregator | standard | $4per Mtok in$24per Mtok out$0.4per Mtok cachedai-gateway.vercel.sh · read 2026-08-25 |
About
GPT Realtime 2.1 — a text and audio model from OpenAI, sold by 3 companies from $4 in and $24 out per million tokens.
It takes text and audio and returns text and audio, with a context window of 128,000 tokens. It was published in July 2026, trained on material up to September 2024. Its sellers say it can reason step by step. The catalogue files it under speak. Three companies sell it. The cheapest is $4 in and $24 out per million tokens at Microsoft Azure AI.
Every current figure
- Maker
- OpenAI
- Register
- model
- Takes
- text + audio
- Returns
- text + audio
- Context
- 128,000 tokens
- Longest answer
- 32,000 tokens
- Published
- July 2026
- Knowledge to
- September 2024
- Licence
- not read
- Sellers
- 3
- Price
- $4 in and $24 out per million tokens — Microsoft Azure AI
Known as 2 names
gpt-realtime-2.1openai/gpt-realtime-2.1