Affiliate disclosure: we may earn commissions when you sign up through some links below, at no extra cost to you. This never affects our pricing data, comparisons, or recommendations. Learn more.

AI Pricing Guru Labs · Video generation

What AI video really costs after failed and unusable footage

List price counts generations. Video Labs counts accepted clips and usable seconds. We run fixed prompts repeatedly, preserve billing and latency evidence, and grade artifacts separately from instruction following.

See measured results Read the test contract Z.ai measurements · Novita partner Prices monitored daily

First measured result

CogVideoX-3 needed 29% more budget than clean per-second math

Z.ai's official $0.20 per-video rate works out to $0.0400 per requested second for this five-second setup. One billed request was not retrievable and four sampled seconds failed the usability rule, raising the observed cost to $0.0516 per accepted second.

Direct requests

8

$1.60 billed

Completion

87.5%

7/8 retrievable

Accepted footage

31s

77.5% of requested time

Effective cost

$0.0516

Per accepted second

Nominal five-second rate$0.0400
Measured accepted-footage rate$0.0516
$0.00$0.03$0.06 / accepted second
Open the full CogVideoX-3 result →

Experiment roster

Measured evidence first; provider claims stay separate

Z.ai

Measured

CogVideoX-3

8 direct API requests · 31 accepted seconds

View evidence →

Z.ai

Next run

CogVideoX-3 stress test

A larger prompt set for motion, camera control, exact text, and repeatability

See the run design ↓

Z.ai

Next run

GLM video understanding

Timestamped extraction and pricing questions from short generated clips

See the run design ↓

Shared test contract

Every provider faces the same buyer-focused accounting

01 · Repeat

Multiple attempts

We repeat each prompt and modality so a single lucky clip cannot decide the verdict.

02 · Preserve

Billing + outputs

Request state, task IDs, latency, output hashes, failures, and observed billing are retained.

03 · Grade

Usable footage

Prompt adherence, temporal consistency, text accuracy, and severe artifacts are scored separately.

04 · Normalize

Accepted-output cost

We report cost per accepted clip and usable second, including billed failures and rejected footage.

Next: larger Z.ai benchmark wave

More generation stress tests plus video-understanding checks

The next runs stay on earned Z.ai bonus credit: a larger CogVideoX-3 prompt set for motion, camera control, exact text, and repeatability, followed by timestamped video-understanding tasks. Results will appear here only after billing evidence, outputs, grading, maintained official price sources, and independent verification are complete.

Check Novita's current video API catalog →

Partner note: Novita is a separate commercial alternative, not the provider behind the Z.ai measurements above. We may earn a commission at no extra cost to you.

Current CogVideoX-3 official pricing and API behavior were checked . Data owner: AI Pricing Guru Data. Re-verification cadence: daily. Benchmark findings describe only the tested settings and prompts.