DATUM

OpenAI releases GPT-Live-1, a two-way voice model for its API

single source · 1 articles · release · confidence: low · first seen 2026-09-10 00:00 UTC

OpenAI has added GPT-Live-1, a voice model, to its API, it said on 10 September. It supports full-duplex conversation — both sides can speak at once, as on a phone call, rather than taking turns — and adds stronger instruction following, custom voices and telephony support. The announcement carries no price, no usage limits, no context window and no evaluation results, and does not say which existing voice models it replaces.

What this means for you

Too early to act on: there is no price, no usage limit and no evaluation for GPT-Live-1, and OpenAI has not said what it replaces. If you build turn-taking voice agents, this is the release to watch — but on today's information there is nothing to migrate to.

Key facts

  • ·OpenAI announced GPT-Live-1 for its API on 10 September 2026. source
  • ·OpenAI describes the model as supporting full-duplex voice conversation, in which both sides can speak at once. source
  • ·OpenAI lists stronger instruction following as a change in the model. source
  • ·Custom voices and telephony support are listed among the model's features. source
  • ·The announcement carries no price, no usage limits and no evaluation results, and names no model it replaces. source

What the sources say

  • OpenAI News — OpenAI's own launch note: one sentence of voice features, with no pricing, limits or benchmarks

Sources

The original reporting. Follow these — they did the work.

← the wire