Skip to main content
BenchLM
Data
Official provider results

Phonic-1 launch evaluation

Phonic’s launch table puts Phonic-1 (high) ahead of GPT-Live-1, Grok Voice 3.0 and Gemini 3.8 Live Extended Thinking on a speech-to-speech index (87.1), τ³-Voice task success (79.9%), multilingual τ³-Voice (83.2%) and Cekura’s live-call reliability (90.2%), with 0.61 s to first audio.

Source snapshots refreshed
Five measurement lanesVOICE / S2S

Reasoning, task completion, conversation dynamics, experience, and latency stay separate.

Back to directory

Benchmark profile

Scope
Speech-to-speech index; τ³-Voice retail, airline and telecom task success; time to first audio; Big Bench Audio spoken reasoning; cost per hour of input audio; Sierra multilingual full-duplex τ³-Voice; Cekura S2S live-call reliability; Podonos blind listener preference
Primary metric
Index score and success rates (higher is better); time to first audio in seconds and cost per hour (lower is better)
Owner
Phonic
Available evidence
Exact values in the launch post’s table image and chart labels

What it measures

Spoken reasoning
Does the system understand and answer spoken requests?
Task completion
Can it follow policy, use tools, and finish a workflow?
Latency
How long do responses, tool calls, and full tasks take?

Available results

Launch comparison

Values as printed in Phonic’s launch table. Index and Big Bench Audio are scores out of 100; τ³-Voice and Cekura are success rates in percent; latency is time to first audio in seconds and cost is US dollars per hour of input audio (both lower is better). A dash means Phonic did not report the cell.

Values as printed in Phonic’s launch table. Index and Big Bench Audio are scores out of 100; τ³-Voice and Cekura are success rates in percent; latency is time to first audio in seconds and cost is US dollars per hour of input audio (both lower is better). A dash means Phonic did not report the cell.
ModelS2S indexτ³-Voice overallRetailAirlineTelecomFirst audio (s)Big Bench AudioCost / hourMultilingual τ³-VoiceCekura S2S pass³
Phonic-1 (high)87.179.972.878.888.30.6198.9$4.2083.290.2
GPT-Live-1 (Astra, medium)83.274.572.576.774.31.3490.1$5.8379.975.6
Grok Voice 3.086.971.666.169.379.50.8198.5$4.80—81.7
Gemini 3.8 Live Extended Thinking (high)82.668.650.072.083.91.3597.7$3.5068.552.4
Cascade (STT + GPT-4.1 + TTS)—31.228.948.016.7————82.9

Blind listener preference

Podonos blind pairwise study run for Phonic. Shares of votes for each side; the remainder were ties.

Podonos blind pairwise study run for Phonic. Shares of votes for each side; the remainder were ties.
ComparisonPhonic-1 preferredOther preferred
Phonic-1 vs GPT-Live-161.5%32.0%
Phonic-1 vs Grok Voice 3.066.5%26.5%
Phonic-1 vs Eleven v4 Turbo78.0%14.0%
Phonic-1 vs human voice actors32.0%57.0%

Interpretation limit

Phonic selected the competitors, settings and figures. It attributes the index, τ³-Voice, latency, Big Bench Audio and cost rows to Artificial Analysis, the multilingual τ³-Voice row to Sierra and the S2S row to Cekura, all dated October 29, 2026; those evaluators’ own pages were not re-checked for every cell, and the third-party figures are not licensed here for Data API redistribution. The Cekura row is pass³ reliability, which matches Cekura’s public board. GPT-Live-1 runs with GPT-6 Astra at medium effort in the Artificial Analysis rows but with GPT-6 Sol at low effort on Cekura. The cascade uses Deepgram Nova-3, GPT-4.1 and Deepgram Aura-1 for τ³-Voice (run by Sierra) and Deepgram Flux, GPT-4.1 and ElevenLabs Flash v2.5 for Cekura. The listener-preference study was run by Podonos for Phonic; the post’s prose rounds the human win rate to about 60% while the chart shows 57.0%. These results do not enter text-model rankings.