Description
Needle 3 is a foundation model from Cactus Compute, built for mobile devices, wearables, robots, smart home devices, cars and microcontrollers. The whole model is a single binary of 8 to 29 MB, based on the company's Simple Attention Network. It is built for three jobs. Tool calls pick the right functions from the ones your app exposes and fill in every argument from what the user said. Structured extraction takes messy text and returns typed fields for a shape you declare, such as an invoice, a booking, a notification or a form. Text embeddings return a vector for a sentence, so an app can search, match and route locally. The company says Needle 3 beats models ten times its size on mobile tool calls and matches models two to three times bigger on extraction, and that it gives up general chat ability to do so. The page lists a browser sandbox where you can try it, and a Python install with pip install cactus-needle.…
Highlights
- Single 8-29 MB model binary
- Tool calls from your app's functions
- Structured extraction into typed fields
- Text embeddings for local search and routing
- Built for phones, wearables and microcontrollers
Ratings & Reviews
No ratings yet
Be the first to rate Needle 3.
People also visit
Needle 3 FAQ
What is Needle 3?
Needle 3 is a small foundation model for tiny devices. It handles tool calls, structured extraction and text embeddings, all in one model of 8 to 29 MB.
How do I try it?
You can try it in the browser sandbox on the site, or install it with pip install cactus-needle.
What is it good at?
The site says it beats models ten times its size at picking tools and filling in arguments on mobile, and matches models two to three times bigger on structured extraction.
Is it a general chat model?
No. The company says it trades general chat capacity to focus on tool calls and extraction.
What are the best Needle 3 alternatives?
Popular alternatives to Needle 3 include Comfyui Triton And Sageattention Installer. See all Needle 3 alternatives





