HeyGen Voice Turbo hits API, free through Nov 30

The leaderboard HeyGen cites in its post is the Controlled Voice arena run by Artificial Analysis. Listeners compare two clips of the same text in blind pairs, and every model speaks with the same eight cloned voices, four US and four UK English, so the comparison is about how well a model renders a given voice rather than whose stock voices sound best. As of October 22, HeyGen Voice Turbo led with an Elo score of 1,202 and HeyGen Voice was second at 1,195, ahead of Qwen's Qwen-Audio-3.1-TTS-Plus at 1,184. The margins are small: the board's 95% intervals overlap, from 1,183 to 1,221 for Turbo, 1,180 to 1,210 for Voice and 1,170 to 1,198 for Qwen, and it lists both HeyGen models with a possible rank of 1 to 3. HeyGen does not appear on Artificial Analysis's separate Provider Voice board, which tests each provider's own native voices and where ElevenLabs' Eleven v4 Turbo led.

Turbo is not a separate voice model. OpenRouter's listing describes it as a lower-latency version of HeyGen Voice with the same 10 default voices, plus voice clones from your own HeyGen account if you supply your key, streaming 44.1 kHz audio in 39 languages with up to 5,000 characters per request; it lists an October 13 release. Artificial Analysis lists Turbo at $4 per million characters against $30 for HeyGen Voice. The offer that the Turbo endpoint is free until November 30 comes from HeyGen's post; no HeyGen page setting out its terms could be found, and OpenRouter's listing, as of October 22, shows a paid rate of $4 per million characters, marked as 50% off $8.

This extends an earlier result. On October 9, Artificial Analysis reported that HeyGen Voice, the first HeyGen text-to-speech model evaluated on that board, had taken first place on the same Controlled Voice board, ahead of Qwen-Audio-3.1-TTS-Plus and Eleven v4 Turbo. Turbo has since been added and, as of October 22, sat just above it at a much lower listed price.