OpenAI Shuts Down Sora — Kling, Runway and Vidu Take Over Video Generation
OpenAI is pulling out of AI video generation. The Sora shutdown announced in March 2026 has been in effect for web and app since 26 April 2026, with the Sora API following on 24 September 2026. The provider that put text-to-video in the spotlight with the original Sora demo in 2024 is now the one leaving the field. The gap is already being filled by Kling, Runway and Vidu — and for teams that built video on an API, this marks a forced migration with a hard deadline.
What actually changed
- Sora web and app have been offline since 26 April 2026. The Sora API still runs but will be discontinued on 24 September 2026 — the move was announced in March 2026.
- The rivals are gaining users. According to Bloomberg, Kling AI grew 4% to 2.6 million weekly active users in the week after the announcement; Runway and Vidu saw gains too.
- Kling was already ahead before the shutdown. In March 2026, Kling reached 7.8 million monthly active users versus 4.7 million for Sora, per Bloomberg.
- On the leaderboard, Kling v3 leads — per the llm-stats video leaderboard with an arena score of ~2075, ahead of Happy Horse 1.0 (2063) and LTX-2 Fast (1936).
What applied before
Since the launch of its app and API, Sora was OpenAI’s flagship in text-to-video. Anyone wanting to build a product or content workflow on generated video had, with the Sora API, a provider inside the same ecosystem as GPT and the other OpenAI models — an argument for teams already standardized on OpenAI.
Despite the visibility, actual usage fell short of expectations. According to research cited across multiple reports, Sora ran at a persistent loss — figures point to high daily compute costs in the millions against comparatively low revenue. Even before the shutdown, Sora trailed Kling in monthly active users.
What applies now
1. Sora is gone as an end product, and the API has an expiry date. Web and app access have been off since 26 April 2026. The API, the last remaining entry point, ends on 24 September 2026. Until then a narrow window remains for migration — after that, there is no Sora endpoint at all.
2. Kling, Runway and Vidu are absorbing the users. Bloomberg reports that Kling AI (Kuaishou) grew 4% to 2.6 million weekly active users in the week after the announcement; Runway and Vidu also gained. Kling already led in March with 7.8 million monthly active users versus 4.7 million for Sora.
3. Quality now sits with the rivals. Per the llm-stats video leaderboard, Kling v3 leads the text-to-video ranking with an arena score of around 2075, followed by Happy Horse 1.0 (2063) and LTX-2 Fast (1936). These are leaderboard/benchmark figures, not our own measurement. Google Veo 3.1 does not appear at the very top of the pure quality ranking but is considered a strong all-rounder: native audio track, lip-synced speech and 4K output in a single pass.
Analysis
The retreat is less a technical than an economic decision. Sora was visible but expensive and lightly used — a combination OpenAI apparently no longer wanted to carry in the race for compute capacity. That the early market leader is vacating the very field it opened so publicly is the real story: AI video is spectacular as a demo but, so far, hard to run as a profitable product.
The specialized providers benefited. Kling, Runway and Vidu are not general-purpose LLM houses but video-focused — and that focus is now paying off, both in user numbers and on the leaderboards. For users this means the question is no longer “Sora or an alternative” but “which alternative fits the use case”. The benchmark lead and native audio strength are split across different models.
What you can do now
If you built on the Sora API: Plan your migration before 24 September 2026. After that date there is no Sora endpoint — switching after expiry is not a fallback, it’s an outage. Also export any Sora assets you still need in time; long-term cloud retention after the shutdown is not guaranteed.
If you need an all-rounder with sound: Google Veo 3.1 produces a native audio track and lip-synced speech in a single pass and delivers 4K — the pragmatic choice when picture and sound should come from one source.
If you’re after a cinematic look and multi-shot sequences: Kling 3.0 leads the text-to-video leaderboards and is built for coherent, multi-part shots.
If you want maximum control over the output: Runway is built for precise control of motion and framing — the choice for teams that don’t just generate but direct.
Entdecke mehr
Three Days with Claude Fable 5 — and Why Opus Stays My Daily Driver
Fable 5 fixed a bug in one shot that Opus failed at twice. I'm still not switching. My honest impressions from the trenches after three days.
GlossarMiniMax
MiniMax is an AI lab founded in Shanghai in 2021 whose open-weight M model family (LLMs) targets very long context windows and low running costs via sparse attention.
LexikonComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.