Cactus Compute Unveils 14MB Needle 2 AI Model

Cactus Compute is targeting on-device artificial intelligence with Needle 2, an open model built for tool calling and structured data extraction. Compact models can shift narrowly defined AI tasks from cloud servers to local hardware, reducing latency, connectivity requirements and computing costs. That approach is particularly relevant for privacy-sensitive applications, intermittent networks and embedded devices that lack the memory or dedicated graphics processors needed by larger models.

The company’s newly released Needle 2 has 45 million parameters but ships as a binary of just 14 megabytes, with a full session requiring 28 megabytes of RAM. Cactus Compute said the model can run across multiple operating systems and constrained hardware platforms. The specifications make offline AI workflows practical on Raspberry Pi computers, wearable devices and other machines without a discrete GPU, including local tool execution and structured information extraction.