- Category
- Developer Tools
- Rank
- No. 3881Tools index
Previous survey · No. 3716 ·
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Data Processing
- GitHub
- 3 stars
- Date
About
Newt is a Swift package that uses an on-device language model to answer multiple-choice, graded, and yes/no questions about a piece of text, returning a probability for each possible answer for use in Apple's Core AI stack. It is inspired by TypeSafe's model "Jev," and its README cautions that outputs should be treated as a ranking rather than calibrated probabilities, since wording changes can shift a score dramatically.
At a glance
Newt is a Swift package that ranks possible answers to a question using a model on the device. It is worth exploring for an Apple app that needs a local suggestion, such as routing a support request to a likely department.
Good fit
Swift developers experimenting with on-device classification who can evaluate their own examples.
Consider something else when
Skip it if you need calibrated confidence scores for automated or consequential decisions.
Start here
Read the published fixtures and compare a clear request with an ambiguous one. In a trial, rephrase the same ambiguous request and check whether the top answer changes before using the ranking in your app.
Cost and requirements
MIT-licensed. The documented Mac setup needs Apple silicon, macOS 27, Xcode 27 and compatible local model resources.
Limits to know
The maintainers show that wording can flip a nearly certain answer. Treat the outputs as rankings. The project reports an iOS build but has yet to verify it on a device.
Sources checked 2026-11-05. Source-based assessment.
Why it made the leaderboard
Newt gives Swift developers a way to get probability-scored multiple-choice, graded, and yes/no answers from Apple's on-device model, filling a real gap for structured, confidence-aware output on iOS/macOS.
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.