Cactus Compute Released Needle 2 for Raspberry Pi 5

The new 45-million-parameter model enables local, offline AI function calling on compact hardware.

Updated on Sept. 24, 2026 in Artificial Intelligence

Isometric editorial illustration showing a detailed view of modular hardware connector pins, representing offline AI integration for compact computing. — Cactus Compute has released Needle 2, a 45-million-parameter AI model enabling local, offline natural-language function calls on the Raspberry Pi 5. AI Illustration. Upload story photo >

Live Poll

Would you prefer using AI tools that function entirely offline on your personal devices?

Cactus Compute has launched Needle 2, a lightweight artificial intelligence model designed to run locally on the Raspberry Pi 5. The model interprets natural-language instructions to execute Python function calls without requiring an internet connection or cloud-based API services.

Why it matters

By eliminating the need for cloud connectivity, this model removes latency and potential subscription costs for developers. It provides a private, offline solution for automating hardware tasks through simple natural language commands.

Needle 2 features a 14MB binary size and utilizes two-bit quantisation. The model achieves decoding throughput of 248-314 tokens per second with a peak memory usage of 46.4MB for the complete Python process.

The players

Cactus Compute

Cactus Compute is a software development organization focused on creating optimized, lightweight artificial intelligence models for edge computing hardware.

Raspberry Pi 5

The Raspberry Pi 5 is a popular single-board computer developed by the Raspberry Pi Foundation that is widely used for education, hobbyist projects, and industrial automation.

The details

Developers can integrate the model by defining functions with the @needle.tool decorator, which allows the system to retrieve up to five tools per request. The architecture employs grouped-query attention and a 256-token sliding context window to maintain efficiency on resource-constrained hardware.

September 24, 2026: Cactus Compute officially released the Needle 2 model.

The Tech Race

The release of Needle 2 mirrors the broader industry shift toward edge AI, where developers prioritize offline processing over reliance on centralized cloud servers. This move challenges larger AI providers by democratizing intelligent automation on low-cost hardware like the Apache 2.0-supported ecosystem.

Users can now automate hardware tasks like switching LEDs or monitoring temperatures using natural language with near-instant response times. The model requires minimal memory overhead, ensuring that other local applications can continue to run smoothly on the Raspberry Pi 5.

The takeaway

Needle 2 demonstrates that sophisticated AI capabilities can function effectively on small, affordable hardware platforms. Developers should consider how moving AI logic from the cloud to the edge can enhance both the privacy and speed of their automation projects.