Cactus Compute introduced a compact needle3 model for mobile devices

Company Cactus Compute released needle3 — a specialized fundamental model designed to work directly on mobile devices. The model's weight varies from 9 to 35 MB, which allows it to be integrated into applications with limited hardware resources without the need to access cloud servers.

The needle3 architecture is optimized to perform three key tasks: invoking external tools, structured data extraction, and creating text embeddings. In internal tests, the model demonstrated a score of 9 points when performing tasks in 1 hour, which, according to the developers, allows it to outperform larger analogs in narrowly specialized usage scenarios.

The model supports flexible customization through fine-tuning and quantization methods, making it easy to adapt to specific business tasks. Due to its minimal size, needle3 provides low latency when processing requests while maintaining high accuracy in extracting information from unstructured data.

The release is targeted at mobile software developers who require the implementation of AI functions with user data privacy compliance. The use of local computing eliminates the transmission of confidential information to external servers, which is a critical requirement for many corporate and consumer applications.