‘AI voice agents get smarter’: OpenAI unveils 3 new models
&imwidth=600&imheight=450&format=webp&quality=medium "‘AI voice agents get smarter’: OpenAI unveils 3 new models")
‘AI voice agents get smarter’: OpenAI unveils 3 new models
OpenAI has introduced three new audio models designed for real-time voice-based applications. The models GPT-Realtime-2, GPT-Realtime-Translate and GPT-Realtime-Whisper are part of the company’s developer platform and aim to improve how software handles live conversations.
OpenAI has introduced three new audio models designed for real-time voice-based applications. The models GPT-Realtime-2, GPT-Realtime-Translate and GPT-Realtime-Whisper are part of the company’s developer platform and aim to improve how software handles live conversations.
The launch marks a shift from basic transcription tools to systems that can listen, understand and respond during conversations. These models are currently available for testing in OpenAI’s developer playground.
What the new models do
The three models are built for different tasks but work together to support real-time voice applications. GPT-Realtime-2 is designed to manage complex conversations. It can handle interruptions, maintain context over longer discussions and connect with external tools to complete tasks. GPT-Realtime-Translate focuses on language translation. It supports translation from more than 70 languages into 13 output languages, making it useful for customer service, education and global communication.
GPT-Realtime-Whisper provides live speech-to-text capabilities. It can generate captions, meeting notes and updates as a person speaks, allowing real-time documentation of conversations.
Move towards real-time AI agents
Trending Stories
&im=FitAndFill=(700,400))
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
)
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
)
)
&im=FitAndFill=(700,400))
)
)
)
With these models, OpenAI is moving towards building AI systems that can act during conversations rather than just respond after receiving input.
This means voice-based applications can:
Understand ongoing conversations
Translate speech instantly
Perform tasks while interacting with users
Such capabilities could improve how businesses use AI in areas like support services, virtual assistants and communication tools.
Early use cases and adoption
Several companies are already testing these models in real-world scenarios. These include Zillow, Priceline and Deutsche Telekom.
These organisations are exploring how real-time voice AI can improve customer interaction, automate processes and make communication more efficient.
Pricing and access details
OpenAI has shared pricing for the new models.
GPT-Realtime-2 starts at $32 per million audio input tokens.
GPT-Realtime-Translate costs $0.034 per minute.
GPT-Realtime-Whisper costs $0.017 per minute.
The models are currently available in a limited testing phase through the developer platform.
What this means for the future
The introduction of real-time audio models highlights how AI is evolving from text-based systems to more interactive tools.
By enabling systems to listen, process and act instantly, OpenAI is expanding the role of AI in communication. As these tools develop further, they may become a common part of everyday applications, from virtual assistants to business operations.
