The cost math — when AI video is viable and when it bankrupts you — step 1 of 6
The unit economics — cost per second across the 2026 lineup
Before anything else, the bare prices. This is API price per second of generated video, before retakes are factored in.
The table
| Model | $/sec | 60-sec spend (raw) |
|---|---|---|
| Sora 2 (720p) | $0.10 | $6 |
| Sora 2 Pro (720p) | $0.30 | $18 |
| Sora 2 Pro (1024p) | $0.50 | $30 |
| Veo 3 (with audio) | $0.40 | $24 |
| Veo 3 Fast | $0.15 | $9 |
| Veo 3.1 Lite (720p) | $0.05 | $3 |
| Veo 3.1 Lite (1080p) | $0.08 | $4.80 |
| Kling 3.0 (standard) | $0.084 | $5.04 |
| Kling 3.0 (Pro, video input) | $0.168 | $10.08 |
| Runway Gen-4 Turbo | $0.05 | $3 |
| Hailuo 02 (768p) | $0.045 | $2.70 |
| Hailuo 02 (1080p) | $0.08 | $4.80 |
| Vidu Q3 | $0.07 | $4.20 |
The four tiers
The lineup separates cleanly into four price tiers:
-
Sub-$0.10/sec — the cheap tier. Hailuo 02, Veo 3.1 Lite, Vidu Q3, Runway Gen-4 Turbo, Kling 3.0 standard. A 60-second piece at raw price runs $3-5. This is where high-volume work (performance marketing, social, B-roll) has to live.
-
$0.10-0.20/sec — the value-quality tier. Sora 2 base, Kling 3.0 Pro. Roughly $6-12 for a 60-second piece at raw price. The sweet spot when you need real model quality but not Sora 2 Pro / Veo 3 premium pricing.
-
$0.30-0.50/sec — the premium tier. Sora 2 Pro (both resolutions). Roughly $18-30 per 60-second piece. Used for hero shots, client work where the output is the deliverable.
-
$0.40/sec — premium with native audio. Veo 3 with audio. $24 per 60-second piece. Used for talking-head / narrated content where the lip-sync and ambient audio matter. (Was $0.75/sec at launch; Google cut prices in September 2025.)
The 10x spread is the strategic question
The cheapest tier is ~11x cheaper than the most expensive tier (Hailuo at $0.045 vs Sora 2 Pro at $0.50). That spread is not "the same product at different prices." It's different products entirely. They serve different jobs.
If your project is "a 30-second product launch trailer for a Fortune 500 client," Hailuo is wrong. The frames will not hold up in a Super Bowl spot.
If your project is "60 short variations for a Facebook ad performance test," Veo 3 is wrong. You'll burn $1,440 on the test set and learn the same thing $300 of Hailuo output would have told you.
Picking the wrong tier — in either direction — is the single most expensive mistake in 2026 video work. The next read covers the retake multiplier that makes all of these numbers worse.
The Higgsfield wrinkle
Higgsfield doesn't price per second. It prices in credits, and the credits buy time on the underlying base models (Sora 2, Veo 3.1, Kling 3.0, etc.). On the Plus plan ($34/mo, 1,000 credits), the effective rate works out to:
- ~14-25 Sora 2 videos a month → ~$1.40-$2.40 per Sora 2 video effective price (much cheaper than raw API).
- ~167 Kling 3.0 generations a month → ~$0.20 per Kling video effective price.
Higgsfield is a price-arbitrage layer on top of being a control layer. If you're going to use Sora 2 a lot AND want camera control, the Higgsfield subscription is probably cheaper than the raw OpenAI API.
But: the credits are usage-capped and the camera-control feature is the actual product. Don't subscribe just for arbitrage — subscribe because you want the deterministic shot vocabulary.
What this gives you
The cost-per-second table is the floor of the planning math. Step 2 turns this floor into an actual budget by adding the retake multiplier — the "generate 10, pick 1" reality that makes every raw price 2-10x more expensive than the table suggests.
The cost math — when AI video is viable and when it bankrupts you — step 1 of 6
The unit economics — cost per second across the 2026 lineup
Before anything else, the bare prices. This is API price per second of generated video, before retakes are factored in.
The table
| Model | $/sec | 60-sec spend (raw) |
|---|---|---|
| Sora 2 (720p) | $0.10 | $6 |
| Sora 2 Pro (720p) | $0.30 | $18 |
| Sora 2 Pro (1024p) | $0.50 | $30 |
| Veo 3 (with audio) | $0.40 | $24 |
| Veo 3 Fast | $0.15 | $9 |
| Veo 3.1 Lite (720p) | $0.05 | $3 |
| Veo 3.1 Lite (1080p) | $0.08 | $4.80 |
| Kling 3.0 (standard) | $0.084 | $5.04 |
| Kling 3.0 (Pro, video input) | $0.168 | $10.08 |
| Runway Gen-4 Turbo | $0.05 | $3 |
| Hailuo 02 (768p) | $0.045 | $2.70 |
| Hailuo 02 (1080p) | $0.08 | $4.80 |
| Vidu Q3 | $0.07 | $4.20 |
The four tiers
The lineup separates cleanly into four price tiers:
-
Sub-$0.10/sec — the cheap tier. Hailuo 02, Veo 3.1 Lite, Vidu Q3, Runway Gen-4 Turbo, Kling 3.0 standard. A 60-second piece at raw price runs $3-5. This is where high-volume work (performance marketing, social, B-roll) has to live.
-
$0.10-0.20/sec — the value-quality tier. Sora 2 base, Kling 3.0 Pro. Roughly $6-12 for a 60-second piece at raw price. The sweet spot when you need real model quality but not Sora 2 Pro / Veo 3 premium pricing.
-
$0.30-0.50/sec — the premium tier. Sora 2 Pro (both resolutions). Roughly $18-30 per 60-second piece. Used for hero shots, client work where the output is the deliverable.
-
$0.40/sec — premium with native audio. Veo 3 with audio. $24 per 60-second piece. Used for talking-head / narrated content where the lip-sync and ambient audio matter. (Was $0.75/sec at launch; Google cut prices in September 2025.)
The 10x spread is the strategic question
The cheapest tier is ~11x cheaper than the most expensive tier (Hailuo at $0.045 vs Sora 2 Pro at $0.50). That spread is not "the same product at different prices." It's different products entirely. They serve different jobs.
If your project is "a 30-second product launch trailer for a Fortune 500 client," Hailuo is wrong. The frames will not hold up in a Super Bowl spot.
If your project is "60 short variations for a Facebook ad performance test," Veo 3 is wrong. You'll burn $1,440 on the test set and learn the same thing $300 of Hailuo output would have told you.
Picking the wrong tier — in either direction — is the single most expensive mistake in 2026 video work. The next read covers the retake multiplier that makes all of these numbers worse.
The Higgsfield wrinkle
Higgsfield doesn't price per second. It prices in credits, and the credits buy time on the underlying base models (Sora 2, Veo 3.1, Kling 3.0, etc.). On the Plus plan ($34/mo, 1,000 credits), the effective rate works out to:
- ~14-25 Sora 2 videos a month → ~$1.40-$2.40 per Sora 2 video effective price (much cheaper than raw API).
- ~167 Kling 3.0 generations a month → ~$0.20 per Kling video effective price.
Higgsfield is a price-arbitrage layer on top of being a control layer. If you're going to use Sora 2 a lot AND want camera control, the Higgsfield subscription is probably cheaper than the raw OpenAI API.
But: the credits are usage-capped and the camera-control feature is the actual product. Don't subscribe just for arbitrage — subscribe because you want the deterministic shot vocabulary.
What this gives you
The cost-per-second table is the floor of the planning math. Step 2 turns this floor into an actual budget by adding the retake multiplier — the "generate 10, pick 1" reality that makes every raw price 2-10x more expensive than the table suggests.