The development

What happened

OpenAI announced API access to GPT-Live-1.1 on 20 November, extending its conversational voice AI beyond the ChatGPT Plus and Pro deployment it dates to 15 October. The developer announcement says the model offers more consistent accents across languages, more precise recognition of letters and numbers, and improved instruction following and task completion; it supplies no numerical comparison in the post. A new standby setting for GPT-Live-1 and GPT-Live-1.1 pauses the model after a configurable silence while keeping a WebRTC call open. The lower standby connection rate does not include backend model or tool usage.

Key facts

  • Developers can select model ID gpt-live-1.1. OpenAI says it has powered voice in ChatGPT Plus and Pro since 15 October; the new announcement concerns API availability, not the first use in ChatGPT.

  • OpenAI attributes improvements in accents, precise spoken details, instruction following and task reliability to the new model. The announcement does not provide a numerical benchmark or independent evaluation.

  • Standby is available for both GPT-Live-1 and GPT-Live-1.1. Developers set a silence timeout from 5 to 60 seconds; the model pauses after that interval while the WebRTC call stays connected.

  • The announcement says conversation resumes when someone speaks again. Standby costs $0.02 per minute versus the standard active rate of $0.05 per minute; backend model and tool usage is billed separately.

  • OpenAI says GPT-Live-1.1 can listen while speaking and delegate complex workflows to a developer-selected backend model without interrupting the conversation’s flow.

The context

Why it matters

TideReach assessment: standby gives voice-app developers a separate cost control for silent portions of an open call, rather than requiring every connected minute to remain active. Its value depends on silence duration and timeout settings, while backend work still adds cost. Teams should test recognition and turn-taking in their own languages and tasks; the announcement alone does not establish a measured quality gain.