GPT Realtime 2

May 2026
GPT Realtime 2
Compare in
on
$32 / $64per 1M audio in / out
1.1 sfirst audio?
30.8τ-Voice
May 2026released

Rankingat minimal effort?

Full ranking →
1Grok Voice 3.094.82Grok Voice 2.094.63Gemini 3.8 Live93.24GPT Realtime 2.191.55GPT Live 190.9
···
9GPT Realtime 1.585.110GPT Realtime 284.711Qwen Audio 3.0 Realtime Flash81.7

Strong at

C
Speech & Voice#10 of 17

Alternatives?

ModelScoreCostPriceWhat changes
Same labGPT Realtimeunstatedsame——

History

4 versions
8486889092Oct 25Jan 26Apr 26Jul 26Release dateSpeech Agent Arena · Task successGPT Realtime1.522.1
8486889092Oct 25Jan 26Apr 26Jul 26Release dateSpeech Agent Arena · Task successGPT Realtime1.522.1

Benchmarks

in Speech & Voice
Big Bench Audio · AA v1.271.8#20
Full-Duplex-Bench · AA subset v1/v1.596.1#7
Time to first audio · AA1.1#15
AA Speech Agent Arena · overall · Elo923.7#14
AA Speech Agent Arena · overall · Task Success84.7#10
AA Speech to Speech Index v2.063.3#12
AA Speech Agent Arena · agentic · Elo920.7#13
AA Speech Agent Arena · non-agentic · Elo912.3#14
τ-Voice · AA · aggregate30.8#10
τ-Voice · AA · airline52#9
τ-Voice · AA · retail21.9#12
τ-Voice · AA · telecom18.4#12
τ³-voice · Sierra · 1.0.0 · standard · retail39.5#3
τ³-voice · Sierra · 1.0.0 · standard · airline56#2
τ³-voice · Sierra · 1.0.0 · standard · telecom20.2#3
τ³-voice · Sierra · 1.0.0 · standard · overall38.5#2

FAQ

Is GPT Realtime 2 free to use?›

There is no free API; the cheapest paid price is $32 per 1M input tokens.

How good is GPT Realtime 2, and what is it best at?›

#10 of 17 in Realtime.

Is GPT Realtime 2 better than GPT Live 1?›

On AA Speech Agent Arena · overall · Task Success, GPT Realtime 2 scores 89.8 vs 90.9 — about the same.

How can I use GPT Realtime 2?›

Through the OpenAI API.

When was GPT Realtime 2 released? Is there a newer version?›

Released 2026-05-07. It is the current version in its line.

Is GPT Realtime 2 open source?›

No, GPT Realtime 2 is a closed model available through an API.

What is GPT Realtime 2?›

GPT Realtime 2 is a closed speech & voice model from OpenAI, released May 2026.