A video model built on living neurons costs half a cent per second

mischief 1 is an image-to-video model built on Wan 2.2, with similar quality and five times faster generation. 1

The software comes from measurements of lab-grown neurons and is applied to existing models, adding less than 0.1% of parameters. 1

The model is available on fal and OpenRouter at about half a cent of a dollar per second. 1

AI-generated content reviewed by the editorial team. How we work

The Biological Computing Co. (TBC), a San Francisco company that uses living neurons to improve AI models, has launched mischief 1, an image-to-video model. The company calls it the first publicly available video model derived from neurons, and says it generates footage up to five times faster than the model it is built on.

The model starts from Wan 2.2, a widely used open-weight video model, and keeps its quality while cutting time and cost. TBC's contribution is the optimization: it grows neurons in the lab, measures how they process information and turns the principles it finds into proprietary software. That software is then applied to existing models. The neurons stay in the lab; customers run software on standard AI hardware that adds less than 0.1% to the base model's parameter count.

On price, fal lists half a cent per second of video. Il 19 ottobre fal aveva presentato FLUX 3 Max, con prezzo scontato fino al 7 novembre 2026. According to the company, a five-second clip therefore costs about two and a half cents, and a hundred clips about two dollars and fifty cents. Those figures make it cheap to try several takes of one idea, an advantage the studio Wonder Studios says it saw in its own workflow. Its co-founder Justin Hackney argues that this efficiency could reshape how films and commercials are made as the approach reaches more advanced models.

The difference from common techniques lies in the source. A giugno Cortical Labs aveva avviato a Singapore un data center con neuroni umani su chip, da 30 watt l'uno. Quantization, distillation and caching are shared industry methods; TBC says its optimizations come from direct measurements of living neurons, which it says makes them hard to replicate. CEO Alex Ksendzovsky names video as the first commercial application and says the company is working on language and physical AI, aiming to build more capable models without simply adding compute.

The model is live today on fal and OpenRouter, five weeks after TBC's collaboration with Amazon Web Services to bring it to enterprise customers.

Related Articles