Breaking

OpenAI Rolls Out Full-Duplex Voice Model GPT-Live-1 and Opens Agents API Public Beta

September 11, 2026 at 04:53 EDT

Foundation Models
AI Agents

OpenAI pushed two developer-facing releases live within the same window: GPT-Live-1, a full-duplex voice model now available in the API at $0.05 per minute, and the Agents API, a public beta that puts the company's Codex harness behind a single managed endpoint. Together the releases extend OpenAI's platform from text and reasoning into always-on voice interaction and long-running autonomous coding-style agents, both aimed squarely at third-party developers rather than ChatGPT's consumer surface.

Sept 10, 2026 · OpenAI

Two Platform Launches: Voice That Listens While It Talks, Agents That Run Themselves

GPT-Live-1 brings full-duplex voice to the API at $0.05/min, while the new Agents API opens public beta access to OpenAI's Codex harness for managed cloud agents.

$0.05

per minute for GPT-Live-1 voice layer

+30 pts

Full Duplex Bench gain vs GPT-Realtime-2.1

80%

fewer interruptions, per Speak's evaluation

hosted sandbox partners for Agents API

Subagent latency: before vs after Agents API (Ciridae)

1×

before

4×

faster after

Ciridae reported a 4x latency reduction on subagent flows and an eval score rise from 0.71 to 0.85 after adopting the Agents API.

Praised

Developers called full-duplex voice "amazing" and welcomed the flat per-minute pricing; one healthcare integrator cut about 23,000 lines of cascaded-pipeline code, roughly an 80% reduction.

Flagged

The voice channel still can't call tools or plugins like text chat can, and the Agents API is limited to US data residency with no zero-data-retention option during the beta.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.