Phonely has released Alma, a language model built specifically for real-time voice agents rather than general chat or coding.
The model is optimized for phone-call behavior: interruptions, spoken identifiers, tool calls and tightly structured conversational flows where response delay is especially noticeable. Phonely offers both hosted and self-hosted deployment options.
The company reports sub-185ms average response latency and 97% function-call accuracy, alongside lower blended token costs than the general models used in its comparison. Those numbers come from Phonely’s own tests and have not been independently reproduced.
The narrow scope is part of the product design. Phonely explicitly says Alma is not intended for creative writing, coding or long-horizon multi-agent work, and recommends pairing it with a broader model when those capabilities are needed.