Spacetime patches and physical simulation
Sora 2 processes video by dividing frames into visual patches and compressing them into a lower-dimensional latent space. This diffusion transformer architecture uses spacetime latent patches to maintain consistency across different resolutions and aspect ratios. The model simulates physical properties like buoyancy and rigidity to ensure objects behave according to real-world laws. If a basketball player misses a shot, the ball will rebound off the backboard instead of teleporting to the hoop. Earlier models often morphed objects or deformed reality to satisfy text prompts, but Sora 2 adheres to established physics constraints. The system produces videos up to 60 seconds long in 1080p resolution. It generates audio, such as dialogue and sound effects, alongside the visual content. The process involves four steps: turning visual data into patches, video compression, spacetime latent patches, and scaling transformers for video generation. The system fragments a video into smaller units called patches, which it binds together in a sequence, using spatial compression to reduce color depth and temporal compression to remove motion vectors. Users can select various aspect ratios including 16:9, 1:1, 9:16, 3:2, and 2:3. Render speeds vary, with 480p being the fastest and 1080p being up to 8 times slower. Users can choose durations of 5, 10, 15, or 20 seconds and pick styles like Film Noir, Pixel Art, or Balloon World. A storyboard feature allows for multi-shot videos with several scenes. To prevent misuse, the system embeds visible watermarks and C2PA metadata in every video. You might have already used ChatGPT to refine these prompts before the tool became less accessible.
The 2026 shutdown and financial pivot
OpenAI stopped the Sora 2 API and consumer applications in 2026. The company decided to shift its focus from entertainment apps to robotics and enterprise tools to prepare for a late 2026 IPO after facing massive server compute costs and a projected $25 billion loss. On September 24, 2026, the sora-2 and sora-2-pro model IDs and the Videos API stopped returning video content. This closure followed the March 2026 announcement that the Sora iOS app and ChatGPT integrations would shut down. OpenAI reported an 83.3% financial burn rate during this period. The company also retired the original Sora 1.0 model in the US on March 13, 2026. Developers received six months of notice before the endpoints failed. The removal included specific model identifiers such as sora-2-2025-10-06 and sora-2-pro-2025-10-06. A failed $1 billion character licensing partnership with Disney contributed to the decision to move away from consumer video, and users must download all their generated clips to a local computer before the servers close because the company will delete all saved videos. The departure from the video market left many creators with broken workflows and no official successor. How will the industry react to the complete disappearance of such a significant model?
Alternatives and current market leaders
Google Veo 3.1 and Kling AI now provide the closest alternatives for creators who need high-fidelity video. Veo 3.1 generates 1080p footage with context-aware audio and allows up to three reference images to keep subjects consistent. Users can set durations of 4, 6, or 8 seconds for these clips. Kling v3 produces clips up to 15 seconds long and allows for up to six scripted shots per generation. Seedance 2.5 produces clips up to 30 seconds long with dialogue, sound effects, and music, but it only reaches 720p resolution. It can use up to 30 reference images and perform lip-syncing for short audio clips. Wan 3 generates clips up to 30 seconds long at resolutions from 480p to 1080p. Gemini Omni 1.1 offers resolutions up to 4K and includes an editing mode to change details in an existing clip. ChatGPT Plus subscribers receive 100 credits per month for 10-second 1080p clips at $25 per month, while ChatGPT Pro users pay $200 per month for unlimited credits and priority generation. Sora 2 is no longer the best choice for creators who prioritize generation speed or temporal consistency.
| Feature | Google Veo 3.1 | Kling AI | Runway Gen-3 |
|---|---|---|---|
| Max Length | 8 seconds | 15 seconds | Not specified |
| Resolution | 1080p | 1080p | Not specified |
| Audio | Yes | Limited | No |




