Introduction

Two weeks after its first in-house voice model topped the charts, HeyGen is back with a faster sibling. HeyGen Voice Turbo (heygen-voice-1-turbo) is a closed text-to-speech tier of the HeyGen Voice 1 model, announced on October 21 and live today on the HeyGen API, on fal, and as ComfyUI partner nodes. There are no open weights. The headline: Turbo now sits at #1 on the Artificial Analysis Controlled Voice Arena with an Elo of 1,202, pushing the original HeyGen Voice down to #2, and a Turbo Free endpoint costs nothing until November 30, 2026.

We covered the original launch in HeyGen Voice 1 tops the Artificial Analysis arena. This piece is only about what's new: the Turbo tier, the free endpoint, and where you can run it.

What Shipped

HeyGen's developer changelog and model docs now list three model ids on the same speech endpoints:

  • heygen-voice-1: the standard model. Works with public, instant and professional voices.
  • heygen-voice-1-turbo: faster speech generation for public and instant voices. Paid.
  • heygen-voice-1-turbo-free: Turbo on shared, limited capacity, free until November 30, 2026. No credit balance needed. HeyGen says response time and availability aren't guaranteed, and you'll get HTTP 429 when the free pool is full.

A few details worth knowing before you switch:

  • Professional clones don't run on Turbo. Turbo only speaks public voices and instant clones. Professional voices stay on heygen-voice-1.
  • Instant cloning is free during the preview. One recording in (HeyGen uses the first 3 minutes), a voice id out, usually ready within seconds.
  • Output: one 44.1 kHz WAV from POST /v3/models/audio/tts, or ordered audio parts as Server-Sent Events from /tts/stream. Each endpoint allows 30 requests per minute per workspace member by default.
  • Standard is on sale too. ComfyUI's announcement says the standard model is 50% off through October 31.

The Numbers

These come from

, which ran Turbo on its blind arena and benchmarks:

  • Controlled Voice Arena: #1 overall at 1,202 Elo, with HeyGen Voice at #2.
  • By category: #1 for Assistants (1,233, ahead of HeyGen Voice at 1,216), tied #1 for Entertainment (1,204, level with Eleven v4 Turbo), and #2 for Customer Service (1,203, behind HeyGen Voice).
  • Pronunciation Robustness: 87.8%, #8 of 33 models, up from HeyGen Voice's 83.2%.
  • Price: $4 per 1M characters, the lowest on that leaderboard. For comparison, Artificial Analysis lists Simba 3.2 at $6.6, Gemini 3.8 Flash-Lite TTS at $11, Voxtral 3 TTS at $13, HeyGen Voice at $30 and Eleven v4 Turbo at $40 per 1M characters.

One honest caveat: the Controlled Voice Arena compares cloned voices reading the same scripts, so it measures how well a model reproduces a voice. HeyGen also notes that the API defaults (similarity: "extra_high" when cloning, expressiveness_boost: 1.0 when speaking) are the settings that took first place. For the closest match to the original speaker, HeyGen recommends similarity: "max" and expressiveness_boost: 0.5 instead.

Where to Run It Today

  • HeyGen playground and API: try public voices at platform.heygen.com and grab an API key there. Start with the public voice Nadine (83548de556df48ba8c09c42a57c51d85), so you don't need a clone.
  • fal: HeyGen Voice and Voice Turbo
    alongside HeyGen Video, with speech in 39 languages and instant voice cloning. The Turbo endpoint is heygen/voice/turbo.
  • ComfyUI:
    : an Instant Clone node (free during HeyGen's preview), a Text to Speech node where you pick Turbo or Standard, and a "HeyGen Voice 1" template that clones and speaks in one workflow.

Try It Yourself

A minimal free request with the public Nadine voice, straight from HeyGen's documented endpoint. Put your key in HEYGEN_API_KEY first:

curl -X POST https://api.heygen.com/v3/models/audio/tts \
  -H "X-Api-Key: $HEYGEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "heygen-voice-1-turbo-free",
    "voice_id": "83548de556df48ba8c09c42a57c51d85",
    "text": "Thanks for your patience. Let me pull up your details now... okay, your parcel left the depot this morning."
  }'

If you get a 429, wait for the Retry-After seconds and try again; a free request never silently switches to a paid model. When you're ready for steady capacity, change the model to heygen-voice-1-turbo and that request is billed.

Conclusion

Turbo is the rare TTS update where the faster, cheaper tier also scores higher in blind listening. If you voice narration, characters or shorts and you're happy with instant clones, the free Turbo endpoint through November 30 is an easy weekend experiment, and the ComfyUI template makes it a drop-in for existing workflows. If you rely on professional clones or need fine controls like speed, pitch and <break> pauses, stay on the standard model for now. Open-weights fans: nothing to download here, it's hosted only.

—Anabel ♡