HeyGen's Avatar 6 builds an avatar from 30 seconds of video
The audio-driven model generates gestures from the delivery of speech, and the company says it topped a third-party blind avatar arena.

Key takeaways
- HeyGen launched Avatar 6 on November 17, 2026, an audio-driven avatar model that generates gestures from the context and delivery of speech instead of replaying one canned motion.
- Users can create an avatar from as little as 30 seconds of video, or connect a YouTube, Instagram or TikTok account to build one from existing footage, HeyGen says.
- HeyGen says Avatar 6 ranked first in New Measure's Avatar Arena, a blind head-to-head vote run by a separate operator; the 84% realism figure comes from HeyGen's own Turing-style test.
- Avatar 6 outputs native 2K video, expands the model's runtime context window from about 3 to about 30 seconds, and upgrades existing personal avatars without new footage.
HeyGen launched Avatar 6 on November 17, 2026, an avatar model that generates gestures from what the script says, the company's fix for the repeated motions that plagued earlier avatars.
HeyGen announced the model in a press release and documents it on its product page and in a technical report. Where earlier avatars reused the same canned gesture, Avatar 6 reads the words and the delivery around them, then moves to match. HeyGen says the model can point, count and turn on cue.
Three specs moved. HeyGen says the runtime context window grew tenfold, from about 3 seconds to about 30 seconds, which lets the model vary motion across a longer stretch instead of defaulting to a loop. Every frame is generated at native 2K (2880x1620). And lip sync improved. Existing personal avatars upgrade to Avatar 6 without new footage.
What it is: HeyGen is a Los Angeles avatar company founded in 2020. It says it reaches more than 40 million users across 196 countries and passed $200 million in annual recurring revenue in June, doubling in eight months.
For a creator, 30 seconds of footage now buys a personal avatar. HeyGen says users can build one from as little as 30 seconds of video, or connect a YouTube, Instagram or TikTok account to make an avatar from footage they already posted. The product page's own setup step asks for at least 15 seconds of talking plus a photo.
Avatar 6 ships in two versions, heygen-avatar-6 and the faster heygen-avatar-6-turbo, through the HeyGen platform and its API. The API adds a motion prompt of up to 1,200 characters. The research page shows single continuous clips of 25 to 28 minutes, and two avatars can share one frame for host-and-guest dialogue.
On the ranking: HeyGen says Avatar 6 placed first in New Measure's Avatar Arena, a blind head-to-head vote run by a separate operator, not by HeyGen. The 84% realism figure is the maker's own. In HeyGen's Turing-style test, 84% of participants rated the avatar as equally realistic as the person it represents.
"People decide in seconds whether a video feels like a person talking to them, and when it doesn't, they stop listening," said Joshua Xu, HeyGen's co-founder and CEO.
The launch follows HeyGen Video, its general-purpose video model, and HeyGen Voice, an in-house voice model the company says debuted at #1 on Artificial Analysis' leaderboard. HeyGen says realistic videos of real people stay inside HeyGen with live identity verification, while third-party platforms can make photo-based and virtual avatars.
Avatar 6 is available today through the HeyGen platform and API, with model IDs heygen-avatar-6 and heygen-avatar-6-turbo; the first video is free and plans start at $29 a month for 600 credits.
Sources
- prnewswire.com - launch, date, context window, 2K, consent model, Xu quote, survey, arena claim
- heygen.com - features, avatar setup (15s/30s), pricing, FAQ, two-avatar conversation
- heygen.com - Turing test numbers, 2880x1620, 25-28 minute durations, consent flow
- developers.heygen.com - model IDs, motion prompt length, aspect ratios, API availability
- avatar.newmeasure.ai - the New Measure Avatar Arena where the #1 ranking comes from