Phonic 1 tops the Speech to Speech Index and Tau Voice agent benchmark

Phonic 1 tops the Speech to Speech Index by the narrowest of margins, and its real lead is on agent tasks. Phonic 1 is Phonic's speech-to-speech model, which takes spoken audio in and answers in spoken audio. Artificial Analysis ranks it first of 22 models with an index score of 87.1, ahead of Grok Voice 3.0 at 86.9, a margin the evaluator itself calls very small. Grok Voice 3.0 had taken the top spot earlier in October.

The index is an average of four measures, and Phonic 1's results across them are uneven. On Tau Voice, which the index uses for agentic performance, it is first at 79.9%, 5.4 points clear of the next model. On Big Bench Audio, the speech reasoning test, it is third at 98.9%, behind StepAudio 3 Realtime at 99.7% and Qwen Audio 3.0 Realtime Plus at 99.2%. The other two measures come from the Speech Agent Arena: there it ranks eighth of 30 on preference, with an Elo rating of 1012, but fourth on task success rate at 92.3%. So seven models rank above it on arena preference, while the agentic test is where Phonic 1 separates itself.

It is also fast. Its time to first audio is 0.61 seconds, fourth overall and the fastest of the top five models in the index. Phonic sells it through its own API at $0.07 per minute, which Artificial Analysis lists as $4.20 per hour of input audio. These results cover only the model's high intelligence setting, tested through the Phonic API. For a team building a voice agent, the ranking supports a closer look at Phonic 1 for task-heavy use; on the overall index it is effectively tied with Grok Voice 3.0.