OpenAI Releases GPT-Live-1 Voice Model via API for Developers

OpenAI on Thursday released GPT-Live-1, a new voice model available through its API, providing developers with a sophisticated tool for building voice-enabled applications and business workflows. The model is engineered for more natural, real-time conversations by being able to listen and speak simultaneously, a key advancement over traditional turn-based systems.

Key Capabilities and Architecture

GPT-Live-1 integrates incoming and outgoing audio processing into a single model, which is designed to eliminate the latency typically associated with separate speech-to-text, language model, and text-to-speech pipelines. The architecture allows the model to delegate reasoning and tool-use functions to other backend models, such as GPT-6 Astra or third-party alternatives.

Key features available to developers include:

• Customization: System prompts can be used to define an agent's tone, pace, and conversational style.
• Robustness: The model is built to manage background noise and periods of silence without interrupting the user.
• Context Retention: It maintains conversational context over extended interactions.
• Telephony Support: The model is equipped for deployment in phone-based voice agent systems, targeting call center and customer service applications.

Performance and Early Adoption

Ad

According to OpenAI, early evaluations have demonstrated significant performance gains. Language learning platform Speak reported that GPT-Live-1 reduced conversational interruptions by nearly 80% compared to its previous systems. In another use case, Tony Stoyanov, co-founder and CTO of a healthcare company, stated the model simplified their codebase by 80% and eliminated 23,000 lines of code.

In benchmark testing, GPT-Live-1 reportedly improved performance on the Full Duplex Bench by 30 percentage points over its predecessor, GPT-Realtime-2.1. When paired with the GPT-6 Astra model, it achieved a first-place ranking on the Tau3 benchmark, which measures the intelligence of voice agents on end-to-end tasks.

Pricing and Availability

GPT-Live-1 is now available in the OpenAI API, with pricing for the front-end voice layer set at $0.05 per minute. Developers will pair this with their choice of a backend model and agent framework, which are priced separately. OpenAI also noted it has expanded its available voice options with new accents and dialects and plans to continue increasing language support.