OpenAI announced on March 24, 2026, that it is shutting down the public Sora API with 30 days’ notice. The reason given internally was blunt: the cost of generating high-quality video at scale was described as “economically irreconcilable” with the pricing that users were actually willing to pay. Generating one minute of Sora-quality video required compute that cost OpenAI multiples of what API customers were paying per request.
The closure affects developers and third-party products that had integrated Sora through the API. The tool itself remains accessible for ChatGPT users across relevant subscription plans — the shutdown is API-only.
Why video generation economics are different
Text generation and image generation have both followed a familiar trajectory: initial costs are high, then hardware efficiency improvements and competition drive prices down to levels that support sustainable API businesses. Video generation has not followed that curve on the same timeline.
The compute requirements for high-quality video — particularly at longer durations and higher resolutions — remain significantly heavier than for comparable image or text outputs. Sora’s quality bar is high, and maintaining that quality at scale means the underlying inference cost per minute of video does not compress the way token costs do for language models.
OpenAI is not alone in facing this. The broader AI video market has been characterized by impressive quality benchmarks and difficult unit economics. Runway, Kling, Pika, and others have all navigated the tension between charging enough to cover compute and pricing accessibly enough to drive adoption.
What this means for developers
Products and workflows built on the Sora API have 30 days to find alternatives or transition to different approaches. The announcement landed in a competitive moment. Runway ML and Kling both offer developer API access and have been iterating aggressively on quality throughout Q1 2026. For teams that need programmatic video generation at scale, both are functional alternatives, though neither matches Sora on every quality dimension.
The irony of the shutdown is timing. March 2026 has been one of the busiest months in AI releases in recent memory, with new models and capabilities announced across nearly every category. The Sora API closure is a reminder that capability and commercial viability are different problems — and that even a genuinely impressive product can fail to find a sustainable price point.
