OpenAI Shuts Down Sora 2 in Its API: Where to Make AI Video Now

OpenAI removed its Sora 2 video models and the Videos API from its developer platform on September 24, 2026. The shutdown covers sora-2, sora-2-pro and all their dated snapshots, and OpenAI's deprecation table lists no replacement. Any app, plugin or automation that generated video through OpenAI's API had to switch to another provider.

What OpenAI switched off

OpenAI's deprecations page is short and blunt. It names the Videos API plus five model names and snapshots, all with the same shutdown date and a dash where the replacement would normally be.

The notice is about the API only. It says nothing about the Sora app, so if you used Sora directly rather than through another tool, check OpenAI's own in-app notices for what applies to you.

OpenAI is not leaving visual media altogether. Its API changelog shows two new image models, gpt-image-2.5-sunburst and gpt-image-2.5-flare, released on September 8, 2026. Video is the part it dropped.

Who feels it

If you pay for an all-in-one creative tool, open its model menu. If Sora 2 has vanished or been relabeled, ask the vendor which model now runs behind your saved templates, because the look and motion of your outputs may have changed.

Where to make AI video now

The good news is that the field is crowded. On our AI model leaderboard, Gemini Omni Flash, FLUX 3 Video and Grok Imagine Video 1.5 share first place on LMArena's text-to-video board, where people pick the better of two clips blind. Sora 2 Pro ranked 11th on that board two days before it was switched off.

Prices are listed per second of finished video. For a fair comparison, here is what the main API options charge for 720p, taken from Google's Gemini API pricing and xAI's price list:

A 15-second ad at 720p therefore costs about $1.50 on Gemini Omni Flash or $6 on Veo 3.1 Standard, before retries. You will rarely keep the first take, so budget for several tries per usable clip.

Your switch-over checklist

A structured prompt makes those comparisons fair, because every model gets the same brief:

Shot: [WHAT THE VIEWER SEES, IN ONE SENTENCE]
Subject: [WHO OR WHAT IS ON SCREEN, WITH 2 OR 3 VISUAL DETAILS]
Camera: [ANGLE AND MOVEMENT, FOR EXAMPLE SLOW PUSH-IN AT EYE LEVEL]
Light and mood: [FOR EXAMPLE SOFT MORNING LIGHT, CALM]
Sound: [DIALOGUE, AMBIENT SOUND OR MUSIC, OR NONE]
Length and format: [SECONDS], vertical 9:16, no on-screen text

What to watch next

• More OpenAI cleanups. The original gpt-image-1 shuts down on October 23 and three more older image models on December 1, 2026. Tools built on them will change again.
• Preview labels. Google still lists Veo 3.1 under preview model names in the Gemini API, so its prices and limits can change with little notice.
• Labeling rules. Realistic AI video published in the EU now falls under the AI Act's deepfake disclosure duty, which has applied since August 2, 2026.

If you want a guided workflow from script to finished ad, our Creators course covers AI video, voice and editing step by step.