LLM Insights中文
← Back to video model comparison
OpenAI logo

Sora 2

OpenAI

Sora 2 costs $0.10 per second of generated video — the 7th cheapest of the 18 video APIs tracked here. A 10-second clip runs about $1.00, and the rate is 2.0× the cheapest option (Veo 3.1 Lite at $0.05/sec), while still coming in 75% under the priciest (Veo 3.1). It outputs up to 720p with a maximum single clip of 12 seconds, and generates a synchronised audio track itself.

Important: the Sora API is scheduled to shut down on 2026-09-24 — not a safe choice for new projects.

Per second

$0.10

Per second

10-sec clip

$1.00

5 sec: $0.50

Resolution

720p

max 12 sec per call

Native audio

Yes

9 peers offer it

Pricing basis

OpenAI meters per second of output; $0.10/sec at 720p.

OpenAI's video model, strong on physical consistency and synced audio — good for social clips and proofs of concept.

Capabilities

Text to videoImage to videoNative audio

Price verified 2026-08-06 · source: official

What it costs on a real workload

ScenarioMonthly costKling 3.0
Social short-form6-second clips, 300 per month$180$151
Ad creative testing10-second clips, 60 variants per month$60$50
E-commerce product loops8-second clips, 1,000 per month$800$672
Side project / prototyping5-second clips, 40 per month$20$17

Alternatives

  • Kling 3.0$0.084/s

    16% cheaper per second than Sora 2.

  • Wan 2.5$0.105/s

    1.0× the per-second rate of Sora 2.

Other models from this provider

Frequently asked questions

How much does Sora 2 cost per video?
Sora 2 is billed per second, normalised to $0.10 per second of output. A 5-second clip costs about $0.50, a 10-second clip about $1.00. OpenAI meters per second of output; $0.10/sec at 720p.
What would Sora 2 cost for a real workload?
At 6-second clips, 300 per month, Sora 2 comes to roughly $180 per month. The same workload on Kling 3.0 would be $151.
Does Sora 2 generate audio?
Yes — Sora 2 produces a synchronised audio track as part of generation, as do 9 other models tracked here.
How long a clip can Sora 2 produce in one call?
Up to 12 seconds per generation at 720p. Vidu Q3-Pro goes furthest at 16 seconds, so longer sequences here mean stitching multiple generations.