OpenAI Quietly Shut Down the Sora in 2026. Here’s What You Missed.
AI video tools moved fast in 2024 and 2025. So fast that most people who heard about Sora when it first appeared in February 2024 — watched the demos, bookmarked a tab, planned to try it — still aren't clear on what Sora is right now in 2026, what changed, what version they'd be using, or how access even works. That confusion is reasonable. The product went through more transitions in 18 months than most software sees in a decade.
This is the guide that puts the whole timeline in one place — what Sora 1 achieved, what Sora 2 improved, what the current status actually is, and why OpenAI's own description of the original model as the "GPT-1 moment for video" is the most important framing you can carry into any conversation about where AI video is heading next.
⚡ Sora in 2026: Current Status at a Glance
Sora is OpenAI's AI video and audio generation model. The original Sora 1 launched publicly in December 2024. Sora 2 — more physically accurate, with synchronized dialogue and sound effects — became the active model. As of April 26, 2026, the Sora consumer app (web + iOS) was discontinued by OpenAI. The Sora 2 API remains available for developers at $0.10–$0.70 per second of generated video, with a scheduled API sunset date of September 24, 2026. OpenAI called the original model the "GPT-1 moment for video" — an explicit acknowledgment that this was the beginning, not the finished product.
Sora 2 delivers physics-accurate motion, world-state persistence across shots, and synchronized audio — a fundamentally different generation of AI video than anything that came before.
How Sora Got Here: The Full Timeline
OpenAI reveals the original Sora model in a research blog post alongside 20 demo clips. The demonstrations — a woman walking through Tokyo, a dog playing in snow with photorealistic physics — shock the industry. OpenAI later calls this "the GPT-1 moment for video."
Sora 1 launches publicly, restricted to ChatGPT Plus and Pro subscribers in the United States and Canada. Includes text-to-video generation, image-to-video, multiple aspect ratios, and the Storyboard feature for multi-shot planning.
Sora 2 releases with significantly expanded capabilities: synchronized dialogue and sound effects, more physically accurate motion, improved controllability, and world-state persistence across shots. API access opens for developers.
Free tier access to Sora ends. Video generation becomes exclusive to ChatGPT Plus ($20/month) and Pro ($200/month) subscribers only.
Sora 1 web is sunset for US users. Only Sora 2 remains active. The Sora 1 platform is fully retired.
The Sora consumer app (web + iOS) is officially discontinued. Sora 2 transitions to API-only access for developers. Consumer subscribers lose Sora access entirely. The Sora 2 API is scheduled for final sunset on September 24, 2026.
Why Sora 1 Was a Genuine Research Breakthrough
The February 2024 reveal wasn't hype — it was a genuine technical leap. Earlier AI video models produced blurry, artifact-heavy clips where objects morphed unpredictably and physical relationships broke down within seconds. Sora 1 was different because of what emerged from scale: object permanence.
In OpenAI's own words, "simple behaviors like object permanence emerged from scaling up pre-training compute." This was the core discovery. The model wasn't explicitly taught that a ball behind a box still exists — it figured that out from training on enough video data. The same emergent understanding applied to lighting consistency, motion physics, and spatial relationships. That's why the demos looked so different from everything that came before.
The original model also supported video generation up to 60 seconds — longer than any comparable system at the time — and offered multiple resolutions and aspect ratios including widescreen, portrait, and square formats optimized for different platforms.
What Sora 2 Actually Changed
Synchronized Audio: The Feature That Redefined the Ceiling
Sora 1 generated silent video. Sora 2 changed that fundamentally. The model generates video and audio simultaneously — not post-processed audio layered on top, but dialogue, background soundscapes, and sound effects generated in sync with the visual content. OpenAI's product page describes it simply: Sora 2 "features synchronized dialogue and sound effects."
For production workflows, this is substantial. An AI-generated clip of a character speaking now produces matching mouth movements and an audio track of that dialogue in a single generation pass. Background sounds — water, traffic, crowd noise — are generated to match the visual environment, not applied from a library.
World-State Persistence: The Physics Achievement Most Articles Skip
The other major Sora 2 advancement is what OpenRouter's technical description calls "world-state persistence across shots." This means that Sora 2 maintains consistent spatial relationships between objects when scenes change. If a cup is on the left side of a table in shot one, it doesn't migrate to the right side in shot two. Objects don't disappear or change shape between cuts.
This sounds basic. It wasn't achievable reliably in any prior AI video model, and it's the technical prerequisite for producing multi-shot content that feels like a real production rather than disconnected AI generations stitched together.
Sora 2 Capabilities — What the Model Supports
| Capability | What It Does | Technical Spec |
|---|---|---|
| Text to Video | Detailed natural language prompts → cinematic video with physics-accurate motion | Sora 2: up to 1080p (Pro API tier) |
| Image to Video | Static reference image → animated video sequence | Maintains source image visual identity |
| Synchronized Audio | Dialogue, sound effects, background soundscapes generated in sync with video | Sora 2 only — not available in Sora 1 |
| World-State Persistence | Consistent spatial relationships and object permanence across multi-shot sequences | Sora 2 architectural improvement |
| Cameo Feature | Insert specific characters or people into AI-generated scenes | Sora 2, subject to content policies |
| C2PA Watermarking | All content carries embedded Content Credentials for AI provenance verification | All tiers, all content — always on |
The Numbers That Define Sora's Story
What Most Sora Coverage Gets Wrong or Skips Entirely
🎬 What the Sora Headlines Almost Never Mention
OpenAI explicitly framed Sora 1 as a "GPT-1 moment" — meaning they knew it was the beginning, not the product. Most coverage treated the February 2024 reveal as if Sora had arrived at its destination. OpenAI's own language in the Sora 2 announcement clarifies the opposite: "The original Sora model from February 2024 was in many ways the GPT-1 moment for video." GPT-1 was a proof of concept. GPT-4 is what most people actually use. The model OpenAI shipped publicly was always Step 1 of a multi-year research program, not the finished line.
Every single Sora video carries C2PA Content Credentials — and SynthID is being added on top. C2PA is a cryptographic metadata standard that embeds origin information — who created the content, when, and with what tool — into the file itself. This travels with the video through downloads, uploads, and re-shares. A journalist, platform, or editor can verify that a clip was AI-generated by Sora even if someone tries to pass it off as real footage. OpenAI additionally announced in May 2026 that Google DeepMind's SynthID — a pixel-level invisible watermark — is being integrated across OpenAI's generation products. Both systems work simultaneously.
The API billing model is per-second of generated video — not per token like GPT. This means a 10-second Sora 2 Standard clip costs exactly $1.00 at API rates. A 10-second Sora 2 Pro clip at 1080p costs $7.00. For production pipelines generating dozens of clips, cost calculation is straightforward but adds up quickly. The per-second model is unique in the AI industry and catches many developers off guard when they first build Sora integrations.
The consumer app discontinuation on April 26, 2026 means most people no longer have access. The headlines that covered this were dominated by speculation about OpenAI's broader strategy, which obscured the practical reality: ChatGPT Plus and Pro users who had Sora access as part of their subscription lost it entirely on April 26. The Sora 2 API remains available to developers with a usage account until September 24, 2026. After that date, no Sora endpoints will accept new requests.
The Red Teaming phase before Sora's launch was unusually rigorous. OpenAI engaged domain experts in misinformation, hateful content, and bias specifically to stress-test Sora before its public release. This resulted in documented safety features being built directly into the model's output pipeline — including content filtering, C2PA watermarking from day one, and explicit refusal policies for categories including non-consensual intimate imagery and content designed to deceive voters. These safety commitments were baked into the architecture, not added as a content layer after the fact.
Current Access and API Pricing
| Access Type | Status (May 2026) | Details |
|---|---|---|
| Free Tier | No Access | Discontinued January 10, 2026. No path to free access currently available. |
| ChatGPT Plus ($20/mo) | Discontinued | Plus lost Sora access when consumer app shut down April 26, 2026. |
| ChatGPT Pro ($200/mo) | Discontinued | Pro lost Sora access when consumer app shut down April 26, 2026. |
| Sora 2 API (Standard) | Active / Sunsetting | $0.10/sec at 720p (Standard tier); $0.05/sec Batch. Sunset: Sept 24, 2026. |
| Sora 2 API (Pro) | Active / Sunsetting | $0.50/sec at 1024p, $0.70/sec at 1080p (Standard); Batch is 50% off. Sunset: Sept 24, 2026. |
Honest Assessment — What Sora Got Right and Where It Fell Short
✅ What Sora Genuinely Achieved
- Established object permanence and physics coherence as baselines for video AI
- First publicly available model with synchronized audio generation (Sora 2)
- World-state persistence across shots — consistent spatial logic
- C2PA Content Credentials built in from day one — responsible by design
- Up to 1080p output quality via API (Sora 2 Pro tier)
- Forced the entire AI video industry to raise its physics accuracy standards
⚠️ Where Sora Fell Short
- Consumer app discontinued April 26, 2026 — most users have no access path
- API sunsetting September 24, 2026 — even developer access is time-limited
- Per-second API billing makes production at scale expensive fast
- No free tier as of January 2026 — never accessible to most of its hype audience
- Rollout was US/Canada-first; global access always lagged the publicity
- The gap between 2024 reveal and public availability created enormous unrealistic expectations
The Bigger Picture — What Sora's Arc Means for AI Video
Sora's trajectory — explosive reveal, methodical rollout, consumer discontinuation, API sunset — follows a pattern OpenAI described themselves with the GPT-1 framing. The first version of a technology establishes proof of concept. Each subsequent version expands capability. Eventually the original infrastructure is retired as the underlying technology moves to a different generation.
The research Sora 1 and 2 produced — on physics-aware video generation, world simulation, synchronized audio — doesn't disappear when the API sunsets. It becomes the foundation for what comes after. For creators and developers evaluating the AI video landscape right now, the question isn't "is Sora still available" — it's "what learned from Sora's architecture is shaping the next generation of video models."
That's the more useful frame for understanding where AI video stands in 2026 and where it's heading in the two years after.
⏱️ Stop Guessing Your AI Video Render Times (and Costs)
With per-second API billing on models like Sora 2, unpredictable render times mean unpredictable budgets. Use our custom calculator to predict exactly how long your next scene will take to render before you commit your credits.
Try the Free Video Render Predictor →Save your time. Never waste another generation credit.
Frequently Asked Questions
Is Sora still available in 2026?
The Sora consumer app — which was accessible to ChatGPT Plus and Pro subscribers — was officially discontinued on April 26, 2026. Free tier access had already ended on January 10, 2026. As of May 2026, Sora 2 is available only through the OpenAI API for developers, with pricing per second of generated video. The Sora 2 API itself is scheduled to sunset on September 24, 2026, at which point no Sora endpoints will accept new requests. OpenAI's Sora 1 was fully retired in March 2026.
What is the difference between Sora 1 and Sora 2?
Sora 1 launched publicly in December 2024 and supported text-to-video and image-to-video generation but produced silent output — no audio. Sora 2, which became the active model in September 2025, introduced synchronized dialogue and sound effects generated alongside the video in a single pass. Sora 2 is also more physically accurate, more controllable, and introduces world-state persistence across shots — consistent spatial relationships between objects that don't change between cuts. Sora 2 also offers the Cameo feature for inserting specific characters into scenes. Both models carry C2PA Content Credentials watermarking on all output.
How much does Sora cost to use via API in 2026?
Sora 2 uses a per-second billing model, which is unique in the AI industry. Standard tier pricing: $0.10 per second at 720p resolution, or $0.05 per second on Batch (with up to 24-hour latency). Pro tier: $0.30 per second at 720p, $0.50 per second at 1024p, and $0.70 per second at 1080p on Standard; approximately half those rates on Batch. A 10-second Sora 2 Standard clip therefore costs $1.00 at API rates; a 10-second Sora 2 Pro clip at 1080p costs $7.00. Access requires an OpenAI API account with at least Tier 2 status. The Sora 2 API is scheduled to sunset September 24, 2026.
What did OpenAI mean when they called Sora the "GPT-1 moment for video"?
OpenAI used this phrase in their official Sora 2 announcement to contextualize the original Sora 1 model within a longer research arc. GPT-1, released in 2018, was the first demonstration that large-scale language model pre-training could produce coherent text — but it was a proof of concept, not a finished product. GPT-4 and its successors are the downstream result of that early work. OpenAI's framing suggests Sora 1's achievement — establishing that physics-accurate video generation could emerge from scaling compute, including object permanence and motion coherence — represents a similar starting point for AI video, not its ceiling. The phrase signals that what's coming after Sora is likely to be as different from Sora 1 as GPT-4 is from GPT-1.
Does Sora watermark its AI-generated videos?
Yes, all content generated by Sora carries C2PA (Coalition for Content Provenance and Authenticity) Content Credentials — a cryptographic metadata standard that embeds verifiable information about the content's AI origin into the file itself. This metadata travels with the video through downloads and re-uploads, allowing journalists, platforms, and viewers to verify that footage was AI-generated even if someone tries to present it as real. OpenAI is a C2PA Conforming Generator Product as of 2026. Additionally, OpenAI announced in May 2026 that Google DeepMind's SynthID — a pixel-level invisible watermark — is being integrated across OpenAI generation products including Sora, adding a second layer of AI provenance detection.
