News
Sora's Sunset: The Cloud Video-Model Reshuffle Studios Need to Plan For
ยท RenderBob team
OpenAI's Sora API retires 24 September 2026. Here is the current cloud video-model landscape and why a studio should never be locked to one.
If you built anything on Sora 2, you have a deadline. OpenAI wound the model down through 2026: the consumer experience closed on 26 April, and the Sora API is set to be discontinued on 24 September 2026, after which the model is fully retired. Anyone with a Sora-dependent pipeline needs to have migrated by then.
The cloud video landscape reshuffled hard this year. The models most commonly cited as current leaders are Seedance 2, Google Veo 3.1, Kling 3, Alibaba Wan 2.7, Grok Imagine Video 1.5, Runway Gen-4.5, Luma Ray 3.2 and MiniMax Hailuo 2.3, and they do not solve the same problem. Public comparisons this year place different models ahead for different jobs: Veo 3.1 for face consistency across scenes and native dialogue audio; Kling 3 for motion quality and cinematic 4K, and often the most credit-efficient; Seedance 2 for unified audio-video realism and image-to-video; Wan for open, cinematic multi-shot work. Runway Gen-4.5, a leader at launch, has slipped down the public leaderboards. Native synchronised audio, exotic a year ago, is now standard across the top tier.
Cost varies enough to matter per shot. One comparison put a 10-second Kling clip around $0.50 and a Veo clip of similar length around $2.50. Choosing the right model per shot, not per project, is the main cost lever, which is why the consensus advice has become to stop picking one model and sticking with it.
Sora's retirement is a concrete illustration of a structural risk: model lock-in. A closed model you depend on can be deprecated on the vendor's timeline, not yours, and take your pipeline down with it. The defences are the same ones this blog keeps returning to: keep prompts model-agnostic so they port cleanly, route per shot rather than committing to one model, and run everything behind an abstraction layer so swapping a provider is a configuration change, not a rebuild. A pipeline that treats each model as a replaceable component survives the reshuffles; one welded to a single vendor does not.
More from the blog
- Visual Dubbing Goes Mainstream: Prime Video Changes the Mouth, Not the Voice
On 9 September 2026, Prime Video launched AI lip-sync for the English dub of Maxton Hall. Human actors record the dialogue. The actors' mouths are regenerated to match.
- Suggestive Editing Arrives: Story-Aware Rough Cuts Move Into the Mainstream
From Eddie AI's story-structured assemblies at NAB 2026 to Premiere's AI Assistant and Resolve 21 search, editing tools are proposing the cut, not only cleaning the footage.