← Capa | Mercado

OpenAI opens GPT-Live-1 full-duplex speech model to developers at $0.05 per minute

2026-11-04 · Fonte:The Decoder

OpenAI opens GPT-Live-1 full-duplex speech model to developers at $0.05 per minute

OpenAI's GPT-Live-1 developer API enables simultaneous speech and listening, with interactivity scores jumping from 45.4 percent to 80.1 percent, though per-minute pricing raises cost questions for builders.

OpenAI has released GPT-Live-1 through a developer API, putting a speech model built for full-duplex conversation into the hands of application builders. Where conventional voice interfaces force a strict alternation between speaking and listening — the model talks, then waits, then responds — a full-duplex design lets a system do both at once, mirroring the overlapping rhythm of natural human conversation. That capability is the core of what OpenAI is now exposing to developers.

The technical progress is quantified in interactivity testing: GPT-Live-1 scores 80.1 percent, up sharply from the 45.4 percent posted by its predecessor — a 34.7-point gain on the metric most directly tied to the quality of live, two-way conversation. A jump of that magnitude indicates that simultaneous talk-and-listen behavior, the defining feature of the model, has improved dramatically between generations.

Pricing will shape the adoption calculus. At $0.05 per minute, GPT-Live-1 is not cheap, and the cost profile bears hardest on precisely the use cases full-duplex unlocks: continuous, free-flowing conversation in which the model may be listening and speaking at any given moment. Developers evaluating the API will need to weigh the measured interactivity gains against a meter that runs for as long as the conversation does.

The release marks the point where full-duplex speech stops being a laboratory capability and becomes a programmable interface. With interactivity performance up nearly 35 percentage points between generations and API access now open to builders, the practical question for the developer ecosystem is shifting: less whether simultaneous talk-and-listen can be made to work, and more which applications generate enough value to justify a per-minute meter running for the full length of every conversation.

Tags: OpenAI · GPT-Live-1 · Speech AI · Full-Duplex · Developer API · Voice AI


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado