Sora 2 API Review: OpenAI Video Generation Features Before 2026 Shutdown

The Sora 2 API from OpenAI offered asynchronous video generation through dedicated endpoints until the service ended on September 24, 2026. Launched in late 2025, it supported text-to-video and image-to-video workflows that produced clips with synchronized audio, dialogue, and sound effects. No official replacement model has been announced for external developers.

Launch Timeline and Core Capabilities

OpenAI released Sora 2 on September 30, 2025, initially through a consumer app restricted to the United States and Canada. API access followed shortly afterward via the Videos API, with endpoints for job creation at POST /v1/videos and status polling or webhooks. The model family included sora-2 at 720p resolution and sora-2-pro supporting up to 1080p output. Typical clip lengths ranged from 4 to 20 seconds, with extensions allowing cumulative lengths near 120 seconds through multiple passes.

Key technical strengths included improved physics simulation, reusable character consistency, and batch job support. Resolutions were fixed at 720x1280 or 1280x720 for the base model, while the pro variant added 1024p and 1080p options. Content policies restricted outputs to material suitable for viewers under 18 and prohibited real individuals, public figures, copyrighted characters, or certain human-face inputs from reference images.

Pricing, Limits, and Post-Shutdown Status

API pricing before closure stood at $0.10 per second for sora-2 at 720p. The sora-2-pro tier charged $0.30 per second at 720p, $0.50 at 1024p, and $0.70 at 1080p, with batch jobs discounted approximately 50 percent. No free tier existed for API usage. Rate limits scaled by organization tier, ranging from roughly 25 to 375 requests per minute depending on the model and usage level. Organization verification was commonly required for higher limits.

The service fully closed on September 24, 2026, after the consumer app and web interfaces had already shut down on April 26, 2026. OpenAI cited internal refocus on agents, coding tools, and robotics as the reason. Existing users could export generated content during a limited window via sora.chatgpt.com/sunset, after which remaining credits transferred to other OpenAI products such as Codex. Third-party aggregators occasionally list legacy access at reduced per-second rates, yet official availability remains zero. Security measures during operation included strict content filters and organization-level verification, though no independent security certifications specific to the video endpoints were published.

Compared with contemporaneous alternatives, the Sora 2 API stood out for native audio synchronization but lacked the persistent character libraries found in some competing platforms and carried higher per-second costs than basic text-to-video services. Its rate limits and content restrictions proved more stringent than several open-source or smaller-provider options that allowed broader input types.