Overview
Sora is OpenAI's flagship video generation model, initially unveiled as a demo in February 2024, and **officially upgraded to Sora 2 on October 1, 2025**, marking the product's transition from a 'demo' to a 'production-grade tool'.
Sora 2 features three breakthrough upgrades over the first generation: **native audio-visual synchronization** (no longer post-production compositing), **Cameos character cameos** (users can insert specified real person/character images into videos), and **real physics engine** (more stable physical consistency).
**⚠️ Important Change (April 25, 2026)**: **OpenAI officially announced the shutdown of the Sora video generation model**, with the **API service to be terminated on September 24, 2026**. The Sora generation entry on the ChatGPT web interface will also be taken offline simultaneously. This marks OpenAI's voluntary exit from the AI video generation track, redirecting resources to the GPT-5.6 series and Codex/Workspace Agents. The AI video track has entered a competitive landscape dominated by ByteDance Seedance, Kuaishou Kling, and Google Veo.
Key Features
- ⚠️ API service termination on 2026/9/24: OpenAI officially announced the shutdown of Sora on 2026/4/25, with less than 3 months of usage window remaining; not recommended for long-term workflows
- Native audio-visual synchronization (Sora 2): Sora 2 can simultaneously output corresponding audio (ambient sound, dialogue, soundtrack) during video generation, eliminating the need for post-production compositing
- Cameos character cameos: Upload an image to have it appear as a character in AI videos, supporting both 2D and real person styles
- Real physics engine: Upgraded physical consistency, with significantly better performance in collisions, fluids, reflections, etc., compared to the first generation
- Top-tier video realism: Details, lighting, and materials approach real-life footage, making it one of the industry benchmarks for video quality
- Deep integration with ChatGPT: One-click invocation via ChatGPT Plus/Pro, seamlessly integrated with conversation scenarios (also being taken offline)
Use Cases
- Transitional video generation tasks before September 24 (not recommended for new projects)
- Comparative reference for migrating existing workflows to Kling / Seedance / Veo / HappyHorse
- Last experience window for researchers and creators to test Sora 2's capabilities
Pros
- Sora 2 once elevated AI video from 'demo' to 'production-grade', with audio-visual synchronization + Cameos + physics engine as benchmark capabilities
- Backed by the OpenAI ecosystem, with image quality, lighting, and materials close to real-life footage
- One-click generation within ChatGPT, low usage barrier
Pricing
**⚠️ API service will terminate after 2026/9/24, please do not start new long-term subscriptions.** Current subscriptions: Free (suspended since 2026-01-10); Plus $20/month (1000 credits per month, 720p, max 5 seconds, with watermark); Pro $200/month (10000 credits per month, unlimited slow speed, 1080p, max 20 seconds, no watermark, includes Sora 2 Pro). API billed per second: 720p $0.10/second, 1024p $0.50/second. Third-party proxies (e.g., API Yi) cost approximately $0.12 (standard) / $0.80 (Pro) per use.
Summary
Sora 2 was once a ceiling for AI video generation, with **audio-visual synchronization + Cameos + physics engine** being nearly irreplaceable in professional scenarios. However, OpenAI **officially announced the shutdown of Sora on 2026/4/25**, and the **API will terminate service on 2026/9/24**, making Sora no longer a long-term option. **For new projects, prioritize: Seedance 2.5 (ByteDance, 30-second one-shot output), Kling 3.0 (Kuaishou, domestic video ceiling), Google Veo 3, HappyHorse (Alibaba, ranked first on Video Arena dual leaderboards)**; the remaining 3-month transition window can be used to migrate existing workflows.
Version History
- Official shutdown announcement (2026-04-25): OpenAI officially announced the shutdown of the Sora video generation model; API service to terminate on 2026/9/24; voluntary exit from the AI video track
- Sora 2 / Sora 2 Pro (2025-10): OpenAI's video model officially productionized: native audio + significant physical consistency upgrade, made available to ChatGPT Pro and API; forms the top tier with Kling 3.x and Veo 3
- Sora 1.0 (Demo) (2024-02): OpenAI's first public demo, with 60-second long video physical realism stunning the industry