Avataar Varya: AI Video at $0.005 per Second

Avataar has launched Varya, an AI video generation model priced at roughly $0.005 per second, about 20 times cheaper than Veo, Runway, Kling, or Luma. Announced June 11, 2026, Varya was built by distilling Alibaba's Wan 2.2 down to a four-step model, producing a five-second 720p clip in 45 seconds on a single Nvidia H200, versus more than 20 minutes for the base model.

Try It: Generate a Clip for Cents

You can generate video from a text prompt or a reference image directly on Varya's website today. At $0.005 per second, a 30-second sequence costs around 15 cents to draft, which makes Varya a practical option for storyboarding and volume testing before you commit budget to a premium model. The model is tuned for Indian festivals, food, clothing, and architecture, so it handles culturally specific scenes that generic video models often get wrong.

Why It Matters for Creators

The headline is the price floor. While most AI video tools charge $0.10 or more per second, Varya resets the cost of a generated second by 20x without dropping to throwaway quality. That changes the economics of iteration: you can afford to generate dozens of variations rather than rationing renders. It is the same pressure that premium video models have been competing on, now pushed to a new extreme by aggressive distillation.

Key Details

Speed: A five-second 720p clip renders in 45 seconds on an H200, roughly 10 times faster than Wan 2.2, by running four inference steps instead of 50.

Price: About $0.005 per second, compared with $0.10 or more for Veo, Runway, Kling, and Luma.

Open weights: Avataar plans to release Varya on India's AIKosh portal with training data included, so developers can self-host or fine-tune it.

What to Do Next

Run a side-by-side test: generate the same prompt on Varya and on your current video tool, then compare cost and quality for your actual use case. If Varya holds up for drafts, route early iteration there and reserve premium models for final renders. Watch the AIKosh portal for the open-weight release if you want to self-host.