{
  "publisher": "Voice Arena",
  "board": "https://voicearena.com/tts-leaderboard/us-english",
  "provenance": "SpeechifyAI's reading of Voice Arena's public US English TTS leaderboard on 16 Oct 2026: the median time to first audio (p50) of every model the board gives a latency reading, from its latency distribution and its leaderboard table, which agree, plus Simba 3.2's quality standing on the same board. The board says its latency is measured from gcp-us-central1 and that run dates vary by model. Since the 24 Sep 2026 reading, Gradium's row on the board is its October 2026 model, Gradium-TTS Real-time, and the board shows the 587.5 ms ElevenLabs reading on its Eleven v3 row, not Eleven v3 Conversational.",
  "metric": "time to first audio, p50, milliseconds, from Voice Arena's automated measurements",
  "measuredFrom": "gcp-us-central1",
  "capturedOn": "2026-10-16",
  "date": "16 Oct 2026",
  "modelsOnBoard": 21,
  "simba32Quality": { "rank": 7, "rankRange": "6 to 10", "elo": 1030, "eloInterval": 6 },
  "latency": [
    { "model": "Gradium Gradium-TTS Real-time", "ttfaP50": 68 },
    { "model": "SpeechifyAI Simba 3.2 Real-time", "ttfaP50": 123, "ours": true },
    { "model": "Inworld Realtime TTS 2 Research Preview", "ttfaP50": 168.5 },
    { "model": "Cartesia Sonic-3.5 Real-time", "ttfaP50": 250 },
    { "model": "Smallest AI Lightning 3.1 Pro Real-time", "ttfaP50": 263.5 },
    { "model": "Fish Audio S2 Pro Real-time", "ttfaP50": 267 },
    { "model": "Fish Audio S2.1 Pro Real-time", "ttfaP50": 283 },
    { "model": "Cartesia Sonic-3.6 Real-time", "ttfaP50": 340.5 },
    { "model": "xAI Grok TTS Real-time", "ttfaP50": 354 },
    { "model": "Microsoft Azure Dragon HD Omni", "ttfaP50": 485 },
    { "model": "Murf AI Falcon 2 Real-time", "ttfaP50": 528 },
    { "model": "ElevenLabs Eleven v3", "ttfaP50": 587.5 },
    { "model": "OpenAI gpt-4o-mini-tts", "ttfaP50": 811.5 },
    { "model": "Google DeepMind Gemini 3.1 Flash TTS", "ttfaP50": 890 }
  ]
}