Murf AI has introduced Falcon 2, a new proprietary voice foundation model designed to deliver more natural speech, lower latency and reduced costs for real-time voice applications.
According to Murf AI, Falcon 2 has achieved higher naturalness scores than OpenAI’s Realtime API and ElevenLabs’ Flash and Turbo models on the independent Artificial Analysis Speech Arena benchmark. The company also claims the model can deliver its first audio in under 100 milliseconds.
Cost is another major focus of the new model. Falcon 2 is priced at $0.01 per minute of generated speech, while ElevenLabs’ Turbo and Flash models cost around $0.05 per minute. At one million minutes of generated audio, Murf estimates Falcon 2 would cost about $10,000, compared with $50,000 for the competing ElevenLabs models.
Murf said Falcon 2 was built around a lightweight architecture that prioritizes inference efficiency instead of simply increasing model size. The company believes this approach can help enterprises deploy voice AI at scale while keeping latency and infrastructure expenses under control.
The Bengaluru-based company is also tailoring Falcon 2 for requirements common in the Indian market. The model supports more than 150 voices across over 35 languages, with broad coverage of Indian languages. It is designed to handle multilingual conversations, code-switching and different accents.
Falcon 2 also includes a semantic processing layer that helps the system understand sentence context and adjust elements such as pacing, emphasis and pronunciation. Its conversational training includes patterns from customer service calls, support interactions and booking-related conversations.
The model further improves text normalization for enterprise use cases involving numbers, currencies, dates, addresses, pincodes and vehicle registration numbers.
For large-scale deployments, Murf said Falcon 2 can handle up to 10,000 concurrent calls while maintaining stable latency. It also provides data residency across 11 regions and supports on-premise deployment. Businesses can additionally create custom brand voices through instant voice cloning using just 10 seconds of recorded audio.
Murf’s testing also highlighted Falcon 2’s latency performance. In tests using a third-party geographically distributed API relay, the company said Falcon 2 ranked first across most of the regions evaluated, with time-to-first-audio remaining below 100 milliseconds.
The launch comes as businesses increasingly explore voice agents for customer support, bookings, commerce and other phone-based services. Murf believes factors such as multilingual support, code-switching, response speed and cost per minute will be critical to making these systems practical at enterprise scale, particularly in India.
Murf AI Co-founder and CEO Ankur Edkie said the company’s focus was not simply on improving speech quality or reducing latency independently, but on developing a system that could improve both while keeping deployment costs manageable.
With Falcon 2, Murf AI is positioning its voice technology as a lower-cost alternative for enterprises looking to deploy high-volume, real-time conversational AI without compromising on speech quality or response speed.
