Table of Contents Show
OpenAI has made GPT-Live-1.1 available to developers through its API, extending a voice model that had already been used in ChatGPT to applications built by other companies. The availability was announced in the OpenAI Developer Community on 20 November 2026, while OpenAI’s API changelog dates the feature entry to 19 November.
What developers can use
The model is identified as gpt-live-1.1. OpenAI describes it as a full-duplex voice model: it can listen while speaking, allowing more natural interruptions and conversational turn-taking than a strictly alternating speech interface. According to the company’s documentation, GPT-Live-1.1 improves speech transcription, pronunciation and accent consistency across languages compared with GPT-Live-1.
Developers can configure a live session to delegate tasks that require reasoning or external tools to a separate backend model. That distinction matters for businesses designing voice agents: the voice interface can handle conversation while other systems perform searches, calculations or operational tasks. It does not mean every backend capability is included automatically.
A standby option for voice sessions
Alongside the model availability, OpenAI introduced a standby configuration for WebRTC sessions using either GPT-Live-1 or GPT-Live-1.1. Developers can set an idle timeout between five and 60 seconds. Once that interval of silence passes, inference pauses while the call remains connected, and the conversation can resume when speech returns.
OpenAI lists a standard active-session price of $0.05 per minute and a standby rate of $0.02 per minute. Those figures are the published voice-session rates, not a complete estimate of an application’s cost. Backend model calls and tool usage are billed separately, and the total will depend on how a service is implemented and used.
Potential uses for businesses
API availability opens up the possibility of integrating the model into customer-support interfaces, multilingual assistance, voice-based product navigation and other interactive applications. For businesses serving customers who prefer speaking to typing, more reliable transcription and accent handling may be relevant. However, these are possible applications rather than independently demonstrated results for any particular deployment.
OpenAI’s documentation says GPT-Live-1.1 supports streaming and function calling, while structured outputs and fine-tuning are not listed as supported features. The model also has its own concurrent-session limits, which developers should check against the limits assigned to their OpenAI organization before planning production traffic.
Availability and context
OpenAI’s announcement states that GPT-Live-1.1 had powered voice in ChatGPT Plus and Pro since 15 October 2026. The new development is its availability for developers to integrate using the API, rather than the first appearance of the model in an OpenAI product.
For teams evaluating voice-based automation, the practical questions remain latency, transcription accuracy for their intended languages, integration reliability, privacy requirements and total operating costs. The published model specifications establish the available features, but do not replace testing with representative users and real workflows.
Sources and documentation
Primary sources: OpenAI Developer Community announcement, 20 November 2026; OpenAI API changelog, entry dated 19 November 2026; OpenAI GPT-Live-1.1 model documentation.








