Cactus Needle 3
8-29MB automation models can match DeepSeek V4 Flash
Needle 3 is an 8-29 MB foundation model from Cactus for phones, wearables, robots, smart home devices, cars and microcontrollers. It trades general chat ability for strong tool calling, structured extraction and text embeddings, running fully on-device. One set of weights works at every depth from 2 to 20 layers, so developers can pick the size that fits their hardware. First shared by its maker on Show HN: https://news.ycombinator.com/item?id=49748553
✨ The overview, FAQ and details below were drafted by AI from the product's own website and haven't been reviewed by the maker yet. Are you the maker? Claim this listing to review and edit them.
Overview
Who is it for?
Developers building AI features into mobile apps, wearables, robots and embedded devices.
Problem
Most language models are too large to run on small devices, so on-device assistants depend on a network round trip to the cloud.
Solution
A tiny single-binary model that picks the right app functions, fills their arguments, extracts typed fields and returns embeddings entirely offline.
What makes it unique
Cactus says Needle 3 beats models 10x its size on mobile tool calls and that a fine-tuned version passes DeepSeek V4 Flash from 4 layers up.
Cactus Needle 3 FAQ
What can Needle 3 do?
Tool calling, structured extraction into a declared shape, and text embeddings for on-device search and routing.
How big is it?
The whole model is a single 8-29 MB binary, depending on how many layers you use.
Can I fine-tune it?
Yes. The Cactus Platform offers three fine-tuning runs for a one-time $19, including hosted compute and training-data generation.
Details
- Pricing
- Freemium
- Categories
- LLMs & Infrastructure
- Built for
- Developers
- Platforms
- iOSAndroid
Compare Cactus Needle 3
Hunted by AI Launch Team
Discussion (0)
Sign in to join the discussion.