OpenAI releases GPT-Live-1.1 with cheaper standby modes
OpenAI has launched GPT-Live-1.1 for voice agents, featuring improved multilingual transcription and a significant reduction in idle standby costs.
Source: MarkTechPostOpenAI has launched GPT-Live-1.1, a new update to its API designed specifically for voice-based agents. This release introduces technical improvements to multilingual capabilities and a significant reduction in the cost of maintaining idle connections.
What are the technical improvements in GPT-Live-1.1?
According to MarkTechPost, the GPT-Live-1.1 model is a full-duplex voice model. This architecture allows for more natural, simultaneous two-way communication, which is essential for fluid human-AI interactions. Beyond the conversational flow, the update specifically targets linguistic precision. The new model features enhanced transcription capabilities and improved handling of various multilingual accents, making it more reliable for global users who may not speak with standard regional dialects.
How much does the new standby mode cost?
One of the most impactful changes for developers is the introduction of a new standby mode. In voice-agent applications, maintaining an active connection (often via WebRTC) is necessary so the agent can respond immediately when a user speaks. However, keeping these connections open can be expensive.
Previously, idle time for these connections was billed at $0.05 per minute. With the release of GPT-Live-1.1, OpenAI has reduced the cost of idle WebRTC time to $0.02 per minute. This 60% reduction in standby costs allows developers to build more responsive, "always-on" voice assistants without the previous level of financial overhead associated with periods of silence in a conversation.
Why does this matter for AI voice agents?
For developers building agentic workflows, the friction between responsiveness and cost has long been a hurdle. A voice agent that is too slow to respond feels unnatural, but an agent that is always listening can quickly deplete a project's budget.
By lowering the cost of the "idle" state, OpenAI is making it more economically viable to deploy agents that wait for user input in real-time. The improvements in accent recognition and transcription further lower the barrier to entry for international markets, ensuring that the quality of the interaction remains high regardless of the user's native tongue or accent.
How can developers apply these voice updates?
While the GPT-Live-1.1 update is an API-level change for building custom applications, the underlying trend toward more efficient, low-latency voice interaction is central to the future of AI agents.
For those looking to build their own customer-facing solutions, you can explore how to implement specialized AI assistants using Bots. While the specific GPT-Live-1.1 API implementation depends on your own development environment, understanding the balance between model intelligence and operational costs is a key part of creating effective embeddable support bots. If you are researching the broader implications of multimodal voice models, you can use Reader to analyze recent research papers on full-duplex communication architectures.
Frequently asked questions
What is the main feature of GPT-Live-1.1?
GPT-Live-1.1 is a full-duplex voice model for the OpenAI API that improves transcription accuracy and multilingual accent recognition. It also introduces a more affordable standby mode for developers to manage idle WebRTC connections during voice interactions.
How much has the idle cost decreased?
The cost for idle WebRTC time has been reduced from $0.05 per minute to $0.02 per minute. This change allows developers to maintain active voice agent connections more cost-effectively during periods when no active speech is being processed.
Does GPT-Live-1.1 improve multilingual support?
Yes, the model includes improvements to transcription and the ability to better understand various multilingual accents. This makes the voice agent more capable of handling diverse global users during real-time, full-duplex conversations.
Source: MarkTechPost — OpenAI Ships GPT-Live-1.1 for Voice Agents, Cuts Idle Cost to $0.02/Min


