| Date | What happened |
|---|---|
| March 24, 2026 | Sam Altman informed staff. OpenAI Help Center notice followed. Wide press coverage March 24-28. |
| April 26, 2026 | Consumer Sora app shut down. Sora.com and the iOS/Android apps stopped accepting new generations. |
| September 24, 2026 | API endpoints (`sora-2`, `sora-2-pro`, and all snapshots) stop accepting requests. Builds depending on the API must migrate before this date. |
The Disney licensing deal — signed roughly three months before the announcement — was effectively unwound when the consumer app closed. Two senior people on the Sora team, Bill Peebles and Kevin Weil, exited OpenAI in April 2026 amid the wind-down.
Why did OpenAI shut down Sora 2?
The shutdown was an economics decision, not a quality decision. Reported numbers from press coverage of the wind-down:
- Burn rate: Approximately $1 million per day in compute costs to serve the consumer Sora app.
- Active users: Peaked near 1 million, then dropped below 500,000 in early 2026.
- Lifetime in-app revenue: About $2.1 million.
The arithmetic is brutal: a $1M-per-day compute bill against $2.1M in lifetime revenue is orders of magnitude apart. Even with API revenue layered on top, the consumer Sora product was a structural loss center.
There was an early warning sign that the math was not going to work. Bill Peebles posted on X on October 30, 2025:
We have been quite amazed by how much our power users want to use sora, and the economics are currently completely unsustainable. we thought 30 free gens/day would be more than enough, but clearly we were wrong!
Bill Peebles, then Sora lead, October 30, 2025
That post was five months before the shutdown announcement. Read in hindsight, it was the company telegraphing what was coming. OpenAI tightened free-tier limits during that window, but the per-user compute cost of high-quality video generation was higher than the price any consumer audience would pay.
What was Sora 2 actually able to do?
The model was genuinely capable. Sora 2’s strengths included:
- Long-for-2025 clip lengths. Sora 2 Pro supported 10-second, 15-second, and 25-second clip durations. The 25-second ceiling was the most generous in the market.
- Native audio. Dialogue, music, ambient sound in the same generation. Earlier than Veo 3 shipped the same capability.
- Multi-shot consistency. The model could carry character and style across cuts without abrupt drift.
- 1792 x 1024 high-resolution output on Sora 2 Pro.
- Disney character library. Through the licensing deal, hundreds of Disney IP characters were available as avatars within the consumer Sora app.
Final API pricing for context (these are the numbers the API is still running on through September 24, 2026):
- Sora 2 standard: $0.10 per second at 720p
- Sora 2 Pro: $0.30 per second at 720p, $0.50 per second at 1792 x 1024
- Maximum 25-second Pro clip: $7.50 (720p) or $12.50 (high-res)
What should you use instead of Sora 2?
Four durable alternatives, each with a different strength.
| Replacement | Pick if you want… | Why over Sora 2 |
|---|---|---|
| Google Veo 3.1 | Best audio polish, Google ecosystem | Native synchronized audio, available through Gemini + Flow + Vertex AI, hard to deprecate |
| Runway Gen-4.5 | Best creator tooling, Hollywood-grade output | #1 on Artificial Analysis benchmark, deep timeline editor, real production credits |
| Kling 3.0 Omni | Character + voice carryover across scenes | 15-second clips, strongest narrative-continuity feature on the market |
| Seedance 2.0 | Lowest cost per second | $0.14 per second, 15-second clips, unified audio-video architecture |
If you had a Sora 2 workflow built around its 25-second single-clip ceiling, no single replacement matches that today. Seedance and Kling tie at 15 seconds; Runway is at 10; Veo is at 8 (with 16-30 second scene extension targeted Q2 2026). For longer pieces, every alternative is now a chained-shot workflow.
For audio, Veo 3.1 is the closest direct replacement — both ship native synchronized audio, both run on a major Western platform, both have credible long-term support. If your concern is the model not being available next year, Veo is the safest bet.
For Disney character access specifically: there is no replacement. The licensing deal was OpenAI-specific and has not been re-signed with any other lab. Plan around licensed IP independently of the model you choose.
What is OpenAI doing with the Sora team now?
Sam Altman framed the wind-down around redirecting compute toward science, posting on X around the announcement: “AI will help discover new science, such as cures for diseases, which is perhaps the most important way to increase quality of life long-term.”
In practice, that meant OpenAI:
- Kept the “Sora” brand for an internal research program. OpenAI is keeping the name alive for a research-only world-model effort, internally referred to as a successor (codename “Spud” per press reporting) aimed at simulating the physical world for science and robotics applications.
- Lost key talent. Bill Peebles (Sora lead) and Kevin Weil exited OpenAI in April 2026.
- Moved away from consumer creative video. No public roadmap for a Sora 3 consumer product. OpenAI’s consumer video presence is paused indefinitely.
The bigger story underneath the shutdown: AI video’s compute economics in 2026 reward labs with their own cloud infrastructure and complementary revenue streams (Google has Search and Cloud; Runway has enterprise contracts; ByteDance and Kuaishou have ad revenue from their consumer platforms). OpenAI’s consumer Sora app was a standalone bet that could not subsidize itself.
Get Smarter About AI Every Morning
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon.
Free forever. Unsubscribe anytime.
Frequently Asked Questions
Can I still use Sora 2 today?
Through the API, yes, until September 24, 2026. The consumer Sora app closed April 26, 2026. Anyone with API workflows still in place should be planning a migration before the September cutoff.
Will OpenAI release Sora 3?
No public roadmap. OpenAI is reusing the Sora name for an internal research program focused on world models for science and robotics — not a consumer product. If a consumer Sora 3 ships, it will be a new bet with different economics.
What about ChatGPT’s video features?
ChatGPT’s Sora integration was tied to the same underlying model and is being wound down on the same schedule. Other ChatGPT capabilities (image generation via DALL-E, Custom GPTs, the o-series reasoning models) are unaffected.
What was the Disney deal?
OpenAI signed a licensing agreement with Disney roughly three months before the shutdown announcement. The deal gave Sora 2 users access to hundreds of Disney IP characters as avatars within the consumer app. It was effectively unwound when the consumer app closed on April 26, 2026. No other AI video lab has signed a comparable deal with Disney as of May 2026.
Was Sora 2 actually good?
Yes. Sora 2 was widely regarded as the most capable consumer-facing AI video model from late 2025 into early 2026, particularly on multi-shot consistency and the 25-second clip ceiling. The shutdown was an economics decision, not a capability decision.
What does this mean for AI video as a category?
It means three things. First, AI video is compute-expensive enough that running a standalone consumer business is hard — you need a complementary revenue stream (Google, Runway enterprise, ByteDance ads). Second, the field has consolidated around four credible providers: Google, Runway, ByteDance (Seedance), Kuaishou (Kling). Third, anyone betting on the technology should expect more consolidation, not less, over the next 12 to 24 months.
Sources
- OpenAI Help Center — Sora discontinuation notice
- The Decoder — Two-stage Sora shutdown timeline
- OpenAI Platform — Sora 2 API model card (still live through Sept 24, 2026)
You might also like
- Google Veo 3.1 Explained — the closest direct replacement for Sora 2’s audio capability.
- Runway Gen-4.5 Explained — #1 on the Artificial Analysis benchmark; the high-end-creator alternative.
- Kling 3.0 Explained — strongest character + voice continuity across scenes.
- Seedance 2.0 Explained — cheapest per-second pricing of the leaders.
- The Real State of AI Video in 2026 — our Special Report Vol. 2, with the full landscape.
- AI Video Generation Guide for Beginners (2026).
- Every AI Model Worth Knowing in 2026.
- AI Glossary — every term in plain English.
AI Summary
The headline: OpenAI announced Sora 2’s discontinuation on March 24, 2026. The consumer Sora app closed April 26, 2026. The API runs until September 24, 2026 — then nothing.
Why: Sora was burning roughly $1 million per day in compute while generating only $2.1 million in lifetime in-app revenue. The economics did not work.
What to use instead: Google Veo 3.1, Runway Gen-4.5, Kling 3.0, or Seedance 2.0 all fill the gap.
What it means: OpenAI is keeping the Sora name alive for an internal research-only world-model program. Public consumer video from OpenAI is paused.
What was Sora 2?
Sora 2 was OpenAI’s flagship text-to-video model, released in 2025. It generated 720p video (up to 1792 by 1024 on the Pro tier), supported multi-shot consistency across scenes, and included native audio — one of the earliest production models to ship dialogue, music, and ambient sound in the same generation. The team was led by Bill Peebles.
For roughly six months from late 2025 into early 2026, Sora 2 was the cultural reference point in AI video. It had the standalone consumer app (Sora.com and Sora mobile), a Pro tier on ChatGPT, an API for developers, and a licensing deal with Disney that put hundreds of name-brand characters into the model as available avatars. It was the model the rest of the field measured itself against.
Then it was over.
When is Sora 2 being shut down?
OpenAI ran a two-stage shutdown. The dates that matter:
| Date | What happened |
|---|---|
| March 24, 2026 | Sam Altman informed staff. OpenAI Help Center notice followed. Wide press coverage March 24-28. |
| April 26, 2026 | Consumer Sora app shut down. Sora.com and the iOS/Android apps stopped accepting new generations. |
| September 24, 2026 | API endpoints (`sora-2`, `sora-2-pro`, and all snapshots) stop accepting requests. Builds depending on the API must migrate before this date. |
The Disney licensing deal — signed roughly three months before the announcement — was effectively unwound when the consumer app closed. Two senior people on the Sora team, Bill Peebles and Kevin Weil, exited OpenAI in April 2026 amid the wind-down.
Why did OpenAI shut down Sora 2?
The shutdown was an economics decision, not a quality decision. Reported numbers from press coverage of the wind-down:
- Burn rate: Approximately $1 million per day in compute costs to serve the consumer Sora app.
- Active users: Peaked near 1 million, then dropped below 500,000 in early 2026.
- Lifetime in-app revenue: About $2.1 million.
The arithmetic is brutal: a $1M-per-day compute bill against $2.1M in lifetime revenue is orders of magnitude apart. Even with API revenue layered on top, the consumer Sora product was a structural loss center.
There was an early warning sign that the math was not going to work. Bill Peebles posted on X on October 30, 2025:
We have been quite amazed by how much our power users want to use sora, and the economics are currently completely unsustainable. we thought 30 free gens/day would be more than enough, but clearly we were wrong!
Bill Peebles, then Sora lead, October 30, 2025
That post was five months before the shutdown announcement. Read in hindsight, it was the company telegraphing what was coming. OpenAI tightened free-tier limits during that window, but the per-user compute cost of high-quality video generation was higher than the price any consumer audience would pay.
What was Sora 2 actually able to do?
The model was genuinely capable. Sora 2’s strengths included:
- Long-for-2025 clip lengths. Sora 2 Pro supported 10-second, 15-second, and 25-second clip durations. The 25-second ceiling was the most generous in the market.
- Native audio. Dialogue, music, ambient sound in the same generation. Earlier than Veo 3 shipped the same capability.
- Multi-shot consistency. The model could carry character and style across cuts without abrupt drift.
- 1792 x 1024 high-resolution output on Sora 2 Pro.
- Disney character library. Through the licensing deal, hundreds of Disney IP characters were available as avatars within the consumer Sora app.
Final API pricing for context (these are the numbers the API is still running on through September 24, 2026):
- Sora 2 standard: $0.10 per second at 720p
- Sora 2 Pro: $0.30 per second at 720p, $0.50 per second at 1792 x 1024
- Maximum 25-second Pro clip: $7.50 (720p) or $12.50 (high-res)
What should you use instead of Sora 2?
Four durable alternatives, each with a different strength.
| Replacement | Pick if you want… | Why over Sora 2 |
|---|---|---|
| Google Veo 3.1 | Best audio polish, Google ecosystem | Native synchronized audio, available through Gemini + Flow + Vertex AI, hard to deprecate |
| Runway Gen-4.5 | Best creator tooling, Hollywood-grade output | #1 on Artificial Analysis benchmark, deep timeline editor, real production credits |
| Kling 3.0 Omni | Character + voice carryover across scenes | 15-second clips, strongest narrative-continuity feature on the market |
| Seedance 2.0 | Lowest cost per second | $0.14 per second, 15-second clips, unified audio-video architecture |
If you had a Sora 2 workflow built around its 25-second single-clip ceiling, no single replacement matches that today. Seedance and Kling tie at 15 seconds; Runway is at 10; Veo is at 8 (with 16-30 second scene extension targeted Q2 2026). For longer pieces, every alternative is now a chained-shot workflow.
For audio, Veo 3.1 is the closest direct replacement — both ship native synchronized audio, both run on a major Western platform, both have credible long-term support. If your concern is the model not being available next year, Veo is the safest bet.
For Disney character access specifically: there is no replacement. The licensing deal was OpenAI-specific and has not been re-signed with any other lab. Plan around licensed IP independently of the model you choose.
What is OpenAI doing with the Sora team now?
Sam Altman framed the wind-down around redirecting compute toward science, posting on X around the announcement: “AI will help discover new science, such as cures for diseases, which is perhaps the most important way to increase quality of life long-term.”
In practice, that meant OpenAI:
- Kept the “Sora” brand for an internal research program. OpenAI is keeping the name alive for a research-only world-model effort, internally referred to as a successor (codename “Spud” per press reporting) aimed at simulating the physical world for science and robotics applications.
- Lost key talent. Bill Peebles (Sora lead) and Kevin Weil exited OpenAI in April 2026.
- Moved away from consumer creative video. No public roadmap for a Sora 3 consumer product. OpenAI’s consumer video presence is paused indefinitely.
The bigger story underneath the shutdown: AI video’s compute economics in 2026 reward labs with their own cloud infrastructure and complementary revenue streams (Google has Search and Cloud; Runway has enterprise contracts; ByteDance and Kuaishou have ad revenue from their consumer platforms). OpenAI’s consumer Sora app was a standalone bet that could not subsidize itself.
Get Smarter About AI Every Morning
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon.
Free forever. Unsubscribe anytime.
Frequently Asked Questions
Can I still use Sora 2 today?
Through the API, yes, until September 24, 2026. The consumer Sora app closed April 26, 2026. Anyone with API workflows still in place should be planning a migration before the September cutoff.
Will OpenAI release Sora 3?
No public roadmap. OpenAI is reusing the Sora name for an internal research program focused on world models for science and robotics — not a consumer product. If a consumer Sora 3 ships, it will be a new bet with different economics.
What about ChatGPT’s video features?
ChatGPT’s Sora integration was tied to the same underlying model and is being wound down on the same schedule. Other ChatGPT capabilities (image generation via DALL-E, Custom GPTs, the o-series reasoning models) are unaffected.
What was the Disney deal?
OpenAI signed a licensing agreement with Disney roughly three months before the shutdown announcement. The deal gave Sora 2 users access to hundreds of Disney IP characters as avatars within the consumer app. It was effectively unwound when the consumer app closed on April 26, 2026. No other AI video lab has signed a comparable deal with Disney as of May 2026.
Was Sora 2 actually good?
Yes. Sora 2 was widely regarded as the most capable consumer-facing AI video model from late 2025 into early 2026, particularly on multi-shot consistency and the 25-second clip ceiling. The shutdown was an economics decision, not a capability decision.
What does this mean for AI video as a category?
It means three things. First, AI video is compute-expensive enough that running a standalone consumer business is hard — you need a complementary revenue stream (Google, Runway enterprise, ByteDance ads). Second, the field has consolidated around four credible providers: Google, Runway, ByteDance (Seedance), Kuaishou (Kling). Third, anyone betting on the technology should expect more consolidation, not less, over the next 12 to 24 months.
Sources
- OpenAI Help Center — Sora discontinuation notice
- The Decoder — Two-stage Sora shutdown timeline
- OpenAI Platform — Sora 2 API model card (still live through Sept 24, 2026)
You might also like
- Google Veo 3.1 Explained — the closest direct replacement for Sora 2’s audio capability.
- Runway Gen-4.5 Explained — #1 on the Artificial Analysis benchmark; the high-end-creator alternative.
- Kling 3.0 Explained — strongest character + voice continuity across scenes.
- Seedance 2.0 Explained — cheapest per-second pricing of the leaders.
- The Real State of AI Video in 2026 — our Special Report Vol. 2, with the full landscape.
- AI Video Generation Guide for Beginners (2026).
- Every AI Model Worth Knowing in 2026.
- AI Glossary — every term in plain English.