Skip to main content

OpenAI's New Voice API: Chat Like a Human for Just 5 Cents a Minute

OpenAI Opens GPT-Live-1 API: 5 Cents a Minute for Human-Like Voice AI

Ever wished your app could chat like ChatGPT? OpenAI just made that a whole lot easier. The company has officially released GPT-Live-1, the same voice model that powers ChatGPT's voice mode, to developers through an API. For $0.05 per minute of front-end voice processing, you can now build full-duplex voice assistants that listen and respond in real time. Complex reasoning? That's handled by a separate backend model, billed at its own rate.

Image

Listen and Speak at the Same Time

Traditional voice AI works in three steps: speech recognition, language model reasoning, and speech synthesis. The system waits for you to finish talking before it even starts thinking. GPT-Live-1 flips that script. It's built on a full-duplex architecture, meaning it can generate voice output while still listening. That makes conversations feel far more natural—it handles interruptions, pauses, and quick back-and-forth exchanges without missing a beat. It even knows when to keep talking, pause, listen, or call a tool.

OpenAI says this design significantly cuts latency and eliminates the awkward connection hiccups common in traditional pipelines. Developers no longer have to juggle multiple models themselves. In tests, GPT-Live-1 outperformed GPT-Realtime-2.1 by 30 percentage points on the Full Duplex Bench, with big gains in turn-taking delay and interactive behavior. Paired with GPT-6 Astra at medium reasoning intensity, it also topped the Tau3 test for end-to-end voice agent tasks.

Complex Reasoning? Hand It Off to the Backend

GPT-Live-1 doesn't do all the heavy lifting. OpenAI split the voice model—responsible for real-time listening and speaking—from the backend model that handles deep reasoning. When you ask something that requires search, complex analysis, or a longer task, GPT-Live-1 passes the work to the backend while keeping the voice conversation flowing. That means developers can pick the right backend for the job: a faster, cheaper model for simple tasks, or a stronger reasoner for tough problems.

In practice, GPT-Live-1 natively supports phone scenarios, making it perfect for full-duplex voice agents like restaurant reservations or customer support. Early adopter Speak, a language learning platform, found that GPT-Live-1 reduced the system's active interruptions of learners by nearly 80% compared to previous round-based systems. The model also offers native voice recognition transcription and response text, supports keyword bias and turn detection, and is optimized for noisy environments. OpenAI expanded the available voices to cover more accents, dialects, and languages. You can even combine GPT-Live-1 with tools like Codex—the voice model takes the task, the backend processes it, and the voice conversation picks right back up.

Key Points

  • GPT-Live-1 API is now open to developers at $0.05 per minute for front-end voice processing.
  • Full-duplex architecture lets the model listen and speak simultaneously, handling interruptions and pauses naturally.
  • 30 percentage point improvement over GPT-Realtime-2.1 on the Full Duplex Bench.
  • Complex reasoning is offloaded to a separate backend model, allowing flexible cost and performance choices.
  • Early partner Speak saw an 80% reduction in system interruptions during language learning.
  • Supports phone scenarios, noisy environments, multiple accents, and integration with tools like Codex.