The news: an independent benchmark just put HeyGen on top
On October 22 to 23, 2026, HeyGen announced that its general-purpose video model, HeyGen Video 1, placed first in a blind text-to-video preference test run by Artificial Analysis, a third-party AI benchmarking firm. The test pitted HeyGen Video 1 against the current field of leading video models without revealing which output came from which tool, which is the closest thing the industry has to an unbiased taste test.
According to the Artificial Analysis text-to-video leaderboard, HeyGen Video 1 holds an Elo score of 1160, plus or minus 15, and currently leads the AA-Video-T2V-Silent v2.1 leaderboard with an Elo score of 1144. Elo scoring works like a chess rating: it's a relative measure built from head-to-head comparisons, not an absolute quality grade, so the number matters mainly in how it stacks up against other models on the same leaderboard.
This is a different story from HeyGen Video's product launch on September 30, 2026. That was about what the tool does. This is about where it now sits competitively, measured by someone other than HeyGen itself.
What HeyGen Video 1 actually is
HeyGen Video is built on MiniMax H3 and post-trained by HeyGen. It generates a whole scene, meaning subject, setting, and sound, from a single prompt, and it accepts three kinds of inputs: text-to-video, image-to-video, or reference-to-video.
Clip length depends on the mode. Text-to-video and image-to-video clips can run 5 to 30 seconds, while reference-to-video is capped at 5 to 15 seconds. Every clip comes with its own audio track covering dialogue, ambience, and sound effects, so there's no separate speech pass or lip-sync step to manage afterward.
For creators juggling a recurring cast or a consistent visual style across episodes, two features stand out: support for up to 12 reference images, videos, and audio clips to keep subjects consistent, and fixed seeds for repeatable shots. Both are aimed squarely at multi-shot, continuity-dependent content rather than one-off visuals.
The pricing picture, side by side
Benchmark rank only matters if the price makes sense for your workflow, and this is where HeyGen Video is positioning itself deliberately as the budget option rather than the prestige one.
On the API, HeyGen Video 1 is listed at $2.40 per minute, according to the Artificial Analysis leaderboard. Broken down by resolution, HeyGen's own pricing page puts the cheapest rate at $0.02 per second at 480p, $0.03 per second at 768p, and $0.09 per second at 1080p. There's no free tier, but the API wallet starts from as little as $5, which is a low barrier for testing before committing budget.
For context on where it sits against the rest of the field: for cinematic hero shots where budget is secondary, Wan 3.0 and Seedance 2.5 currently sit at the top of the blind-vote leaderboard and can generate full 30-second takes. HeyGen Video is explicitly positioned as the budget-friendly, photorealistic alternative to those higher-end, higher-cost options.
Why this matters if you're optimizing for cost per video
The practical takeaway here isn't that HeyGen Video 1 is definitively the best model for every job. It's that an independent, blind benchmark just confirmed it's competitive at the top of the field while charging noticeably less per second than some rivals. That combination is rare enough to be worth a re-test if you've settled into a routine with a different tool.
This matters most for a specific kind of creator: anyone making short-form, dialogue-heavy content with a recurring cast, where the cost of a single polished hero shot matters less than the cost of generating dozens of consistent shots across an episode. Fixed seeds and reference-to-video inputs are built for exactly that use case, since they reduce the number of expensive retries needed to keep a character looking and sounding the same from shot to shot.
It's also worth remembering that rankings like this move fast. Kuaishou introduced an early-access preview of Kling 4.0 Flash on September 28, 2026, with the full flagship Kling 4.0 launching globally in October 2026. The leaderboard HeyGen Video 1 just topped is a snapshot, not a permanent state, and the brief doesn't indicate how long HeyGen expects to hold the top spot or how Kling 4.0's full release might shift the rankings.
What to actually do with this information
If you're currently paying premium per-minute rates for a video model mainly because you assumed higher price meant higher quality, this benchmark is a reasonable prompt to question that assumption. The data suggests the correlation between cost and blind-test preference isn't as tight as it used to be, at least on this one leaderboard at this one point in time.
A sensible next step is a direct comparison on your own content rather than trusting the leaderboard alone: generate the same prompt, same reference images, and same shot list in HeyGen Video 1 and whatever model you're currently using, then compare both cost and output quality side by side. The $5 minimum wallet makes that test cheap to run before you decide anything.
If you're building out a multi-episode series inside a tool like Episodo, this kind of pricing and consistency comparison is worth doing at the planning stage, before you've locked in a model across dozens of shots, since switching mid-series is far more disruptive than switching before you start.