Days after OpenAI shut down Sora 2 on March 24, 2026, Sam Altman confirmed that OpenAI has completed pre-training of its next-generation model, internally codenamed Spud. The name immediately sparked speculation: is this the Sora replacement? Is OpenAI about to relaunch video? The short answer is no — and understanding why matters for anyone making decisions about their AI video workflow right now.
What is OpenAI Spud?
Spud is OpenAI's next-generation language and reasoning model. It completed pre-training in March 2026, and Sam Altman has confirmed it is close to release. Based on available information, Spud is a multimodal model — meaning it likely handles text, images, and possibly other input types — but it is fundamentally a language and reasoning system, not a video generation product.
The codename "Spud" is an internal designation. OpenAI has not announced what the model will be called publicly, what its release date is, or what specific capabilities it will ship with. What Altman has confirmed: pre-training is complete, the model is near release, and it represents a significant step forward from OpenAI's current generation of models.
What Spud is NOT
This is the part that matters most for video creators. Spud is not:
- A video generation model — Spud is a language/reasoning model. It does not generate video.
- A Sora replacement — There is no announced OpenAI video product to replace Sora 2. Spud is not it.
- Sora 3 — No Sora 3 has been announced. As of March 2026, OpenAI has not indicated one is in development for consumer or commercial release.
- Evidence that OpenAI is returning to consumer video — The Spud announcement is a language model milestone. It says nothing about OpenAI's video roadmap.
The conflation of Spud with video is understandable — the timing of the Sora shutdown and the Spud confirmation in the same news cycle made them feel related. They are not. They are two separate announcements that happened to coincide.
Why OpenAI killed Sora for Spud
The connection between Spud and Sora's shutdown is not strategic — it is computational. OpenAI is redirecting the GPU resources that were running Sora video generation to Spud and to its enterprise Codex products. This is the primary operational reason for the Sora shutdown timeline: not that video is no longer interesting, but that training and running a frontier language model at Spud's scale requires the infrastructure that was allocated to Sora.
This tells you something important about OpenAI's internal prioritisation. When forced to choose between maintaining a consumer video product and running a next-generation language model, OpenAI chose the language model without apparent hesitation. Video generation lost the compute allocation fight at a company whose foundational competency is language models.
The broader strategic context
The compute reallocation is the surface-level explanation. The deeper context is that OpenAI is re-centering on enterprise AI, reasoning, and robotics infrastructure. The Sora consumer app had lost 66% of its downloads in three months. The standalone consumer video product was not generating the engagement OpenAI needed to justify its infrastructure cost.
Spud — as a language and reasoning model — is more directly aligned with OpenAI's enterprise contracts, the ChatGPT product, and the agentic AI use cases OpenAI is currently competing to capture. It is a better use of GPU capacity, from OpenAI's perspective, than a video app with declining engagement.
What OpenAI's video research is actually doing
OpenAI's video research team has not disbanded — but it has redirected. Based on reporting from Bloomberg and other sources covering the shutdown, the team is pivoting to world simulation research: using video generation technology to help AI systems learn physical logic for robotics applications.
This is a meaningful distinction. World simulation is about training robotic systems to understand how physical objects behave — not about producing visually compelling footage for human viewers. The underlying technology overlaps, but the goal and application are entirely different from consumer video generation.
What this means practically: OpenAI is not quietly working on a better Sora to release later. The team that built Sora is now working on something with fundamentally different objectives. A return to consumer video would require rebuilding the product focus from scratch.
What this means for the video creator competitive landscape
With Sora 2 gone and no replacement announced, the frontier AI video landscape in March 2026 is:
- Kling 3.0 — Kuaishou's model has become the community consensus benchmark for photorealistic cinematic video. Strong physical realism, native audio, complex narrative scenes. Currently the model most comparable to what Sora 2 Pro was doing at its best.
- Veo 3.1 — Google DeepMind's model leads on audio integration and prompt precision. Best for content where sound design matters and where detailed creative briefs need to be executed reliably.
- Runway Gen-4 — Competitive on workflow tooling and iteration speed. Quality ceiling is below Kling 3.0 and Veo 3.1, but strong for high-volume creative testing.
The departure of Sora from the competitive market does not create a quality gap — Kling 3.0 and Veo 3.1 have both advanced significantly since Sora's launch. It creates an access question: where do you get these models, at what cost, and without managing multiple subscriptions.
Should video creators wait for Spud or a future OpenAI video model?
No — and this is not a close call.
Spud is a language model. It will not help you generate video. If it is eventually multimodal in ways that include video understanding, that is a different capability from video generation. It does not fill the Sora-shaped hole in a video production workflow.
A future OpenAI video product — if one ever ships — is not on any announced roadmap. The team is working on robotics world simulation. The infrastructure is being redirected. There is nothing concrete to wait for. Any future OpenAI video product is, at best, years away and entirely speculative.
The practical answer for video creators is to work with what exists now. Kling 3.0 and Veo 3.1 are genuinely excellent models. Kling 3.0 in particular is, by community consensus across AI video forums and creative communities, currently the highest-quality AI video model available. Waiting for a hypothetical OpenAI product while these models exist is not a rational production decision.
An honest take on what Spud will actually deliver
Spud will likely be impressive. OpenAI's language and reasoning models have set consistent benchmarks, and if Altman's confidence about the release timeline reflects the model's actual quality, Spud could represent a meaningful step forward for language, coding, and reasoning capabilities.
For video creative work, however, Spud is irrelevant in the near term. It may eventually integrate with video workflows in the way that GPT-4o integrates with image generation — as an intelligent layer on top of generation models rather than as a generation model itself. That would be useful but is speculative.
The honest summary: Spud is OpenAI's next big bet on language and reasoning. It is probably going to be a significant model. It has nothing to do with video generation. If your primary concern is AI video for marketing and creative production, Spud does not change anything about your workflow today.
What to do right now
If the Sora shutdown disrupted your video production workflow — or if you are evaluating AI video models for the first time — the current recommendation is straightforward:
- Use Kling 3.0 as your primary model for photorealistic and narrative video content
- Use Veo 3.1 when audio integration and prompt precision matter most
- Do not wait for Spud — it is not a video product
- Do not wait for Sora 3 — it has not been announced and there is no evidence it is in development
- Access both models from a single platform to avoid the operational overhead of multiple subscriptions
The AI video quality available today — from Kling 3.0 and Veo 3.1 — is higher than what Sora 2 Pro was delivering at its best for most use cases. The models that can replace Sora for brand and marketing creative exist and are accessible now. There is no strategic reason to pause production waiting for an OpenAI announcement that may not come.