Desert Ant Labs: local, fast models that run on device

Desert Ant Labs, a European artificial intelligence lab, is launching a series of opinionated on-device intelligence models. The company believes that the best way to create efficient intelligence is by building models that operate on-device. These models are small, specialized, and perform tasks such as audio, vision, and text analysis with millisecond response times and no associated costs.

The first 18 models are already available, accessible through a unified SDK for Swift, Kotlin, and JavaScript. Desert Ant Labs is located in Europe, where on-device intelligence is considered the default approach. The company's first product, Detail, initially relied on cloud APIs for features like auto-editing and audio enhancement, but these services became expensive as the app grew in popularity.

Desert Ant Labs decided to train their own models and integrated them into Detail, resulting in faster and cheaper performance compared to cloud-based alternatives. The first 18 models are now live, with full specifications and benchmarks available on desertant.com/models and Hugging Face. Each model can be used for free up to 100,000 monthly active devices, with no tokens, logins, or costs.

By focusing on local models, Desert Ant Labs aims to change the way products are built, leveraging the computational power available in consumers' devices. The company's future plans include creating larger, more capable models and integrating them into the developer experience, providing a seamless workflow for building features that run entirely on-device.