By OpenAI’s Videos API goes dark on September 24, 2026. That’s 16 days. When it does, every request to sora-2, sora-2-pro, and all associated model snapshots returns a 410 Gone. What makes this different from a typical deprecation: OpenAI’s own migration table leaves the “recommended replacement” column blank. There is no Sora successor on OpenAI’s API. Every path forward leads to a different vendor.
What’s Actually Shutting Down
The full deprecation list for the Sora API shutdown:
videos(Videos API endpoint)sora-2sora-2-prosora-2-2025-10-06sora-2-2025-12-08sora-2-pro-2025-10-06
All Sora account data gets deleted on shutdown with no recovery window. OpenAI sent developer notices in March 2026, which means teams have had six months — yet migration rates suggest many haven’t moved. If you’re reading this, check now.
Why OpenAI Killed Video Entirely
This wasn’t a quality call. Sora’s estimated operating cost ran to roughly 5 million per day, while its total lifetime revenue — not monthly, cumulative — came to approximately .1 million. Every 10-second clip cost OpenAI around .30 in compute. Fidji Simo, OpenAI’s applications CEO, told employees the company couldn’t afford “side quests.” Video was the biggest one.
A planned billion licensing deal with Disney for generative character access collapsed — Disney reportedly learned of the shutdown less than an hour before the announcement. That’s how fast the economics overwhelmed the strategy. OpenAI is now firmly agentic. The video chapter is closed.
Your Migration Options
There are four practical paths for the Sora API migration depending on your use case and volume:
Google Veo 3.1 — Best Quality
Veo 3.1 is the quality standard right now. Pricing runs $0.03 per second for 720p lite (no audio) up to $0.60 per second for 4K with audio. It supports native audio generation and is available via Google AI Studio, the Gemini API, and Vertex AI. The catch: it’s not OpenAI-compatible, so you’ll need to adopt the Google SDK. That’s a real migration effort, not a two-line swap. See Veo 3.1 full pricing breakdown for tier details.
Seedance 2.0 — Best for High-Volume APIs
For workloads that generate a lot of clips at controlled cost, Seedance 2.0 is the current front-runner. Pricing: $0.074–$0.092 per second at 480p, $0.161–$0.199 per second at 720p. Audio is included at no extra charge. Access is through BytePlus internationally, or via aggregators like Atlas Cloud using an OpenAI-compatible API format — which means minimal code changes.
Kling 3.0 — Best for Longer Clips
If your use case needs clips beyond 10 seconds — up to 15 seconds natively — Kling 3.0 is the option. It starts at $0.112 per second at 1080p, supports multilingual generation, and offers multi-shot capabilities. Available directly through the Kling API or through aggregator platforms.
Wan 2.7 — Open Source, Self-Host
Alibaba’s Wan 2.7 is a 14-billion-parameter open-source model with a commercially permissive license. Weights are on HuggingFace. If you have the hardware — 24GB VRAM is the practical minimum (RTX 3090/4090/5090) — you run it for free. At scale, that math beats any API. The official self-host guide covers setup end to end. For teams without suitable GPUs, Alibaba Cloud offers API access at $0.10 per second.
The Fastest Migration Path: Two Lines of Code
If you’re using an OpenAI-compatible aggregator (Atlas Cloud, APIMart, AI.cc), the actual code change for your Sora API migration is minimal:
from openai import OpenAI
# Before: pointed at OpenAI
# client = OpenAI(api_key="your-openai-key")
# After: point at aggregator, swap model name
client = OpenAI(
api_key="your-aggregator-key",
base_url="https://api.atlascloud.ai/v1"
)
response = client.videos.generate(
model="seedance-2.0",
prompt="Your existing prompt",
duration=10
)The request and response shapes are close enough that the rest of your integration typically requires no changes. For teams that want to preserve their existing SDK patterns, this is the path of least resistance. Read the Apiyi migration reference for model aliases and parameter differences across providers.
Provider Comparison
| Provider | Price/sec | Audio Included | OpenAI-Compatible | Best For |
|---|---|---|---|---|
| Veo 3.1 (Google) | $0.03–$0.60 | Yes | No (Google SDK) | Highest quality output |
| Seedance 2.0 | $0.074–$0.199 | Yes | Via aggregator | High-volume API workloads |
| Kling 3.0 | $0.112+ | Optional | Via aggregator | Clips up to 15 seconds |
| Wan 2.7 | $0 self-host | Yes | Self-hosted | Privacy and cost control |
| Runway Gen-4 | ~$0.60/clip | Yes | No | Creative/editorial teams |
16 Days. Act Now.
September 24 is a hard cutoff, not a rolling deprecation. There’s no grace period after shutdown, and deleted data doesn’t come back. If the Sora API is anywhere in your stack — even as a secondary pipeline — test your replacement now. Pick your provider, run a clip, confirm your output format. The official OpenAI discontinuation notice lists the exact model aliases being removed.
OpenAI’s video exit isn’t a pause. It’s a permanent product decision. Build accordingly.
