A Practical Proposal for the Sora 2 Team: Release Open Weights Before the Product Sunsets OpenAI’s official discontinuation notice is clear: the Sora web and app experiences were discontinued on April 26, 2026, and the Sora API is scheduled to be discontinued on September 24, 2026. That creates a serious gap for creators, researchers, and small studios who have already built workflows around the model. This post is a respectful proposal, not a complaint. The idea is simple: before Sora 2 disappears from the hosted product stack, consider releasing an open-weight or self-hostable version of the model family, or a carefully scoped subset of it, so the creative ecosystem does not lose access entirely. Sora has already shown that video generation can be more than a novelty. For many users, it became a tool for concept art, previsualization, storyboarding, prototyping, social video, and rapid creative iteration. When a tool like that is removed from public availability, the loss is not only technical. It also affects workflows, communities, learning, and the momentum of independent creators. A practical open-weight release would not need to be a full one-size-fits-all dump of the entire production stack. It could follow a staged model family approach: - Sora 2 Lite for local experimentation and lower-memory systems - Sora 2 Fast for quick previews and lightweight iteration - Sora 2 Pro for stronger fidelity and more demanding workloads - Sora 2 Turbo for optimized inference paths - Sora 2.x updates for measured improvements over time, such as v2.1, v2.2, v2.3, v2.5, and eventually v3 That kind of release strategy would help preserve continuity. It would let creators keep using Sora locally at home or on a studio machine, while allowing OpenAI to continue focusing on higher-level hosted offerings, safety research, and future frontier systems. The strongest argument for this approach is not sentiment. It is continuity. Creative tools do real work. People build habits, libraries, prompt systems, style references, editing pipelines, and production schedules around them. If a model is shut down without a migration path, the result is immediate disruption. Independent artists, small teams, educators, and experimental creators are usually the first to lose access. A self-hostable or open-weight Sora would also reduce pressure on centralized infrastructure. If some users can run the model locally, the hosted service is no longer the only bottleneck. That matters for cost, scalability, and long-term resilience. It also aligns with a broader industry trend: OpenAI has already released open-weight models under a permissive Apache 2.0 license, including gpt-oss models intended for broad use and self-hosted deployment. That precedent shows that open weights are not incompatible with OpenAI’s technical direction. This proposal is not asking OpenAI to give away everything in a way that removes oversight or responsibility. It is asking for a balanced release model. OpenAI could define usage boundaries, provide safety guidance, publish recommended hardware requirements, and ship regular minor updates. That would be enough to preserve the creative value of the model while still respecting operational realities. There is also a strategic benefit. A controlled open-weight release would build trust. It would show that OpenAI recognizes not only the value of product launches, but also the value of long-term access for the people who adopt and support a model early. Creators remember when a platform helps them. They also remember when it disappears. If Sora 2 must sunset as a hosted service, then an open-weight or locally runnable version would be the most constructive outcome. It would keep the model alive in the hands of the community, prevent a hard stop in creative workflows, and allow Sora’s technical legacy to continue evolving outside a single subscription or API endpoint. In short: if the hosted product must end, the weights should not. Let the creative ecosystem keep building, testing, improving, and sharing. That is how a model becomes more than a product. It becomes infrastructure for imagination.

Welcome to the community! I’ve seen some really impressive work with open weight image and video generation recently. Of course, Sora and Sora 2 were incredibly impressive when they came out, and it’s always a shame when a model is lost (I’m still crying for gpt4-0314 and gpt 4.5), but I do suspect sora has been superseded at this point. Have you tried alternative models?

Yeah, I wonder if you could even run Sora 2 on consumer hardware?

Proposed Local Sora 2 Model Family As part of this proposal, OpenAI could consider preserving the existing Sora 2 model variants as locally deployable releases. For example, the following versions could be made available as self-hostable/open-weight packages, subject to appropriate licensing and safety requirements: • sora-2 • sora-2-pro • sora-2-2025-10-06 • sora-2-2025-12-08 • sora-2-pro-2025-10-06 These versions could serve different purposes and hardware requirements. The dated releases could preserve specific model snapshots, while the standard and Pro versions could represent the main local deployment options. OpenAI could then continue the family with optimized releases such as Sora 2 Lite, Sora 2 pro Lite, Sora 2 , Sora 2 pro, Sora 2 Turbo, Sora 2 pro Turbo, and future Sora 2.x versions, allowing users to choose between generation quality, speed, model size, and required VRAM. The goal would not be to make every version run on an ordinary consumer PC. A Pro model could require a high-end workstation or a multi-GPU home server, while smaller variants could be optimized for more accessible hardware. This would make local Sora deployment a practical spectrum rather than a single hardware requirement. Most importantly, preserving these model snapshots would give creators the option to continue using the specific Sora 2 behavior and capabilities they have already incorporated into their workflows, rather than forcing everyone to migrate immediately to a completely different model.

Thanks for the thoughtful proposal. The web/app experience ended April 26, and the API shutdown is scheduled for September 24: https://help.openai.com/en/articles/20001152. We haven’t verified an open-weight or self-hosted Sora release. We’ll pass along the continuity request; we don’t have a release plan or timeline to share.

Release Sora 2 open weights before shutdown

API Deprecations

Topic		Replies	Views
Open Sora 2 Local Checkpoints for Model Continuity Feedback sora	2	169	November 5, 2026
Save the Sora 2 API and please do not discontinue Feedback sora-2	1	385	September 6, 2026
Please continue sora or its api or launch another video generation app Feedback api , sora	2	388	October 7, 2026
A Case for Open-Sourcing o4-mini: Empowering the Next Generation of Local Reasoning Open Models	0	178	February 15, 2026
Open ai, please keep sora ai, please cancel the shut down Feature requests sora	10	3659	April 1, 2026

Release Sora 2 open weights before shutdown

Related topics