OpenAI shuts down the Videos API and every Sora 2 model snapshot on 2026-09-24, and its deprecations page lists no recommended replacement. If you have a product calling sora-2 or sora-2-pro, you have under a month to move, and the honest answer is that no single model replaces it. You pick a replacement per use case.
TL;DR
- Deprecation announced 2026-03-24, shutdown 2026-09-24. Affected: the Videos API,
sora-2,sora-2-pro,sora-2-2025-10-06,sora-2-2025-12-08,sora-2-pro-2025-10-06 - OpenAI's own deprecations page leaves the replacement column empty. There is no drop-in successor
- Closest by behavior: Veo 3.1 for prompt adherence and native audio, Gemini Omni Flash for cheap iteration with audio, Kling v3 for cost-controlled volume, Seedance 2.5 for long single-pass clips
- Sora is still selectable on the 8frame canvas at 56 credits for a 4-second clip. Our access ends when the provider's does, on the same date as everyone else's
What actually got deprecated
OpenAI notified developers on 2026-03-24 that the Videos API and the Sora 2 aliases and snapshots were deprecated, with a shutdown date of 2026-09-24. The list covers both the standard and Pro models, including the pinned snapshots teams use precisely so that behavior does not change under them.
The empty replacement column is the part that matters for planning. When OpenAI deprecated older model families it usually pointed at a successor. Here it does not, so "wait for the OpenAI version" is not a migration plan.
The migration table
Per-clip prices below are what the 8frame canvas charges, credits at $0.01 each in a top-up pack. Pick by what you were using Sora for, not by which model sounds closest.
| What you used Sora 2 for | Move to | Price | Why |
|---|---|---|---|
| Prompt-faithful hero shots with audio | Veo 3.1 Standard | 448 cr / 8s with audio, 224 silent | Strongest reading of shot vocabulary of anything we host |
| Cheap iteration with sound in the loop | Gemini Omni Flash | 51 cr / 3s, 135 cr / 8s | Audio in the same pass, 3 to 10 second range |
| High volume where unit cost decides | Kling v3 Standard | 86 cr / 5s with audio, 57 silent | Per-second billing, audio optional |
| Clips longer than 8 seconds | Seedance 2.5 | 157 cr / 5s at 720p | Durations to 30s, native audio |
| Budget drafting at 720p | Veo 3.1 Lite | 54 cr / 8s with audio, 33 silent | Cheapest Google tier, same prompt behavior as its siblings |
| Physical realism on a budget | Hailuo H3 | 88 cr / 5s at 2K | 5 to 15 second range |
Sora 2 itself, while it lasts: 56 credits for 4 seconds, 112 for 8, 168 for 12. Sora 2 Pro: 280, 560, 840.
The migration checklist for teams shipping on the Videos API
Map every call site first. Not just the generation call. Anything that parses the Videos API response shape, stores an OpenAI video id, or assumes Sora's duration options (4, 8, 12 seconds) is part of the surface area. Duration is where most migrations break: Veo generates at 4, 6, or 8 seconds, Kling takes any integer from 3 to 15, and Omni takes 3 to 10.
Test parity on two vendors, not one. The reason to pick two is not redundancy for its own sake. It is that a single provider deprecating a model with no successor is exactly the failure you are currently living through.
Re-check your audio assumptions. Sora generated audio with the clip. Veo and Kling make audio optional and price it separately, roughly doubling the per-second rate when on. Gemini Omni always generates it. If your product promised sound, the cheapest replacement is not the cheapest silent tier.
Budget for the price change. Sora 2 standard at 4 seconds was 56 credits. The closest replacements land between 54 and 135 credits for a longer clip, so per-second cost is broadly comparable, but a naive "same duration, same budget" assumption will be wrong in both directions depending on the model.
What we are doing about it
We host Sora on the canvas today and we are moving off it before the date, same as everyone else building on the API. The reason the canvas is 20-odd models rather than one is precisely this: a model retiring is a picker change, not a rewrite. That is worth saying plainly rather than pretending we saw it coming.
FAQ
Is the Sora app also gone? This article is about the developer API, which OpenAI's deprecations page dates to 2026-09-24. Check OpenAI's own product pages for the consumer app status rather than trusting a third-party summary of it, including ours.
Will OpenAI ship a Sora 3 API? Nothing has been announced. The deprecations page lists no successor, so plan on the assumption that there is not one.
What is the closest single replacement? For most teams, Veo 3.1 for the shots that must be right and Gemini Omni Flash for everything else. See Gemini Omni vs Veo 3.1 for how those two split the work, and Sora 2 alternatives for the wider field.
Migrating off one API is a good moment to stop depending on one. Run Veo 3.1, Kling v3, Gemini Omni Flash, and Seedance side by side on the same prompt on 8frame, from $19/month.