OpenAI shut down Sora in March this year. A product whose monthly active users fell short of a tenth of its expectations had burned through resources for too long and could no longer sustain itself.
But the AI video track hasn't cooled down; it has only become more chaotic. At least four players are now vying for the position Sora vacated.
Who is running now
Runway Gen-4, the choice most recognized by professional creators this year. Released in March 2025, it focuses on character consistency — you only need to provide a reference image, and it can maintain the visual consistency of the same character across different lighting, scenes, and angles. Anyone who has worked on ad production will understand how important this is: the biggest pain point of AI video used to be that the same character would completely change appearance when the scene shifted.
Maximum single generation length is 20 seconds, with native 4K 60fps output. Pro and Enterprise accounts with 200 clips per month cost roughly $100.
Kling 2.0, under Kuaishou, launched its international version in January 2026. It costs 60% of Runway's price and generates faster (Runway takes 30-90 seconds, while Kling delivers a full version in 5-10 minutes). 200 clips per month cost about $56-64.
The latest Kling 2.6 also adds native audio generation — AI synthesizes sound effects and ambient audio directly while generating video. If this feature is perfected, it could eliminate an entire post-production workflow.
Google Veo 3 is deeply integrated with the Google ecosystem. It can be used directly in YouTube Studio, Google Workspace, and Google Ads, saving teams already living in Google products the hassle of importing and exporting. But if you are not a Google ecosystem user, its appeal is much weaker.
Seedance is the only option that is open source and comes with a commercial license. It can be self-hosted and fine-tuned for specific needs, making it suitable for enterprises with data security requirements or teams looking for differentiated training.
What the scores say
A ten-prompt benchmark (covering product ads, narrative shorts, and abstract content) yielded these results:
| Tool | Overall Score |
|---|---|
| Runway Gen-4 | 8.8/10 |
| Kling 2.0 | 8.4/10 |
| Google Veo 3 | 8.0/10 |
| Seedance | 7.6/10 |
The gaps are small, but Runway has a clear advantage in temporal consistency and motion control — camera cuts are stable and physical effects are more believable.
An interesting workflow
Professional creators have developed a combined approach: use Runway Gen-4 to generate character consistency images, then import them into Kling for animation.
The logic is: Runway's Reference system is best at locking in character appearance (up to three reference images can be uploaded), while Kling performs better on dynamic physical effects, with faster speed and lower cost. Each tool does what it does best.
This dual-tool workflow has become a de facto standard among social media creators.
Pricing models have changed
Before Sora shut down, the mainstream in the AI video space was still subscription-based credit systems. Now it is shifting toward per-second billing: you pay for the number of seconds of video you generate.
This is more friendly for teams with predictable usage — you can calculate how much a project will cost without budget overruns from opaque credit deductions. For platforms, it incentivizes users to optimize prompts and reduce generation volume rather than blindly stacking output.
This pricing change may seem minor, but it redefines the business logic of AI video tools — from subscription software to pay-as-you-go infrastructure.
What Sora's failure shows
OpenAI used its strongest generation capabilities for Sora, but made the wrong commercial choices: high prices, strict quotas, and a fragmented user experience. Meanwhile, Kuaishou's Kling, with lower prices and more generous generation limits, pulled away a portion of users.
The deeper issue is: AI video is a tool highly dependent on creative workflows, not a product users can buy and use independently. Sora tried to enter this market with a closed product logic, but professional creators need modules that can be embedded into their workflows, not an isolated box.
Runway and Kling survived primarily because they positioned themselves as open tools, not closed platforms.
Sources: CocoLoop, After Sora: Best AI Video Generators 2026 Complete (Digital Applied); Runway Gen-4 Maximum Clip Length 2026 (AI Tools Digest); Moonshot's Kimi K2.5 is 'open,' 595GB, and built for agent swarms (VentureBeat)