One-minute AI avatar test: Griffin Turing call vs a 60 s Sume clip

A 60-second Sume avatar clip costs $11.04 at standard, $14.70 at plus and $33.00 at max with no product shown, plus $0.95 once to create the avatar. Tavus ran its Griffin-Lite Turing test after a one-minute video call, and 48% (26 of 54) of participants judged the avatar to be human (Tavus: Griffin, read 2026-10-05). Both are one minute long, but one is a live call and the other is a video you watch, so they test different things.

The Sume rates are on the API pricing page, read 2026-10-05. Tavus's page gives no price for Griffin-Lite, which is not available to customers, so there is no cost to compare against.

Two different minutes

Put the two minutes next to each other and the difference is the viewer's role.

The comparison is not about which is better. A live call asks whether a person can be fooled in a minute of conversation, and a clip asks whether the message works for the people watching. Tavus's one-minute call has a participant who is part of it; your clip has an audience who is not. The cost line is the other difference: one has a published price per second, and the other does not have a price on its page.

The cost of one minute

Compute the cost once and reuse it. Show one minute at each tier, with and without a product, and add the $0.95 creation fee.

The output shows, for each tier, one minute without a product, with a product, and the total including the avatar fee. At standard that is $11.04, $11.64 and $11.99; at plus $14.70, $15.48 and $15.65; at max $33.00, $34.80 and $33.95. Note that the last column adds the creation fee to the no-product price, so it is for a clip without a product.

from decimal import Decimal as D

rates = {"standard": ("0.184", "0.194"), "plus": ("0.245", "0.258"), "max": ("0.55", "0.58")}
for tier, (plain, product) in rates.items():
    one = D(plain) * 60
    withp = D(product) * 60
    print("%-8s $%.2f  product $%.2f  with avatar fee $%.2f" % (tier, one, withp, one + D("0.95")))

Running your own one-minute test

A 60 second clip is also the longest a single request can be, since the estimated duration window is 4 to 60 seconds. So a test of one full minute is the largest unit you can buy in one job. If you want to test how viewers react, run the clip and a disclosure line, and ask them afterwards what they thought they watched.

Keep the test small and cheap. Render the clip at standard to judge the script and pacing, and move to plus or max only for the version you intend to publish. A 60 second test at each tier would cost $11.04 plus $14.70 plus $33.00, or $58.74, and is rarely necessary since preview stills are tier-independent and can answer the look question first.

Do not borrow the 48%

Do not copy Tavus's number onto a rendered clip. The 48% is about a live call, with sample sizes of 54 and 41 that leave wide intervals, as shown in the sample size post. A clip viewer is a different task. If you publish an avatar clip, say plainly that it is synthetic, as covered in the disclosure post.

More

For a wider cost view, see the price ladder and the side by side comparison.