Phonely launches Alma, a voice AI model trained on over 10 million phone conversations
SiliconANGLE Kyt Dotson ● Covered by 2 sources
Phonely launched Alma, a voice AI model trained on 10 million real phone calls. It’s built to sound less awkward on calls and answer faster than GPT-4.1.
Based on reporting by SiliconANGLE, Kyt Dotson — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Phonely has rolled out Alma, a large language model built specifically for voice agents and trained on more than 10 million real phone conversations. That’s the headline feature here: not text, not synthetic scripts, but actual call traffic. The company is betting that this matters because so many voice bots still sound a little wrong when people talk to them on the phone.
The pitch is partly about timing. Phonely says Alma reaches its first token in under 185 milliseconds, while GPT-4.1 is roughly 500 milliseconds. In voice, those gaps are not academic. A beat too long and the call starts to feel robotic, like the system is fishing for a line instead of holding a conversation.
Phonely is also pushing the cost angle hard. It says Alma costs 55 cents per blended million tokens, which it claims is 84% cheaper than GPT-4.1 at $3.50 per million and 90% cheaper than GPT-5.4 at $5.63. The company also says Alma is 63% faster on first token than GPT-4.1 and 82% faster than GPT-5.6.
But speed is only part of the argument. Phonely says Alma keeps improving by learning from where conversations break down, using that as a feedback loop so the model can adjust sooner instead of waiting for the next training cycle. It also says the system works with existing transcribers and text-to-speech providers, which should make it easier for teams to plug in without rebuilding everything around it. And because Alma is already answering millions of Phonely calls each month, the company is framing this as a model that learns from live use, not just a lab bench.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of voice AI product: trained on voice, built for voice, priced for voice. Too many teams are still trying to paste a chat model onto a phone line and hoping the pauses won’t give the game away. Alma sounds like a direct shot at that habit, and honestly, it’s about time.
Read more about this at: SiliconANGLE