The most humbling story in AI video is not about a new model — it is about the old famous one. Sora, the product that made text-to-video a household phrase, was discontinued: OpenAI announced the shutdown on March 24, 2026, the Sora app went dark on April 26, and the API is being wound down over the course of 2026, as documented in Sora's reference history and confirmed by every major field guide published since. October's roundups describe the market fine without it.

Veo 3.1: the quality reference

Google's Veo 3.1 (released October 2025) has settled in as the field's quality benchmark in the current generation of comparisons — 720p-and-up output with the strongest overall fidelity scores across the guides we checked. Google is also reported to operate a second, complementary video model (referred to in technical coverage as Gemini Omni), giving it unusual depth in the modality. If your criterion is simply "the best-looking generated footage available today," the published guides agree on Veo first.

Runway Gen-4: control as the product

Runway has taken a different bet: Gen-4 competes on controllability rather than raw fidelity. Its documented strengths are director-style tooling — granular camera movement controls, consistent characters, and editing workflows aimed at people who produce video for a living rather than generate clips for social feeds. For commercial and narrative work where "close enough" is not close enough, that control surface is the reason creative teams keep it in the stack.

The challengers: lip-sync and open weights

Kling 3 has differentiated on lip-synced dialogue — generating characters that speak specific lines convincingly, which matters for marketing and localization. And MiniMax's open-weight H3 model has become the reference point for teams that need to fine-tune video generation on their own footage — the open-source lane in a market of hosted products.

What Sora's disappearance teaches

Sora's shutdown — a research milestone that lost the product race — is worth internalizing before you build anything on a single vendor's video API. Video generation is expensive to run, fast-moving, and strategically volatile; the guides that survived Sora all make the same practical recommendation: keep workflows tool-agnostic, keep prompts portable, and for anything durable, prefer platforms with an API and an export path.

Frequently asked questions

Is Sora gone for good?

The app was shut down on April 26, 2026, and OpenAI said the API would follow later in 2026. There has been no announcement of a successor product.

Which AI video model is best now?

Published October 2026 guides put Veo 3.1 first on output quality, Runway Gen-4 first on creative control, and Kling 3 first on lip-synced dialogue — "best" depends on which of those your project needs.

Can I self-host an AI video model?

MiniMax's H3 is distributed with open weights, making it the current reference option for teams that need to fine-tune and run video generation on their own infrastructure.

Sources