# Jeff releases small decision models for local use

_Published Monday, September 28, 2026 at 9:09 PM EDT · AI · Latest · Tier 2 — Notable_

![Fine-tunes of Qwen3.5 and Gemma 4 for zero-shot classification - firelex/jeff — Primary](https://opengraph.githubassets.com/54a5ab9fdb1c0aed3f686fb7bac17a62462cad6829c02c5674159a35da733ff9/firelex/jeff)

The Jeff project has released model weights and code for small AI models that choose among options described in plain language and return a probability for each choice. Its published models include 0.8B and 2B parameter versions based on Qwen3.5, plus a Gemma 4 variant. Developers can run them locally through an interface compatible with Jev requests.

The project reports a 22 ms decision time on an RTX PRO 6000 and 28 ms on an Apple M4 Max for its 0.8B model. It says the models work best for classification and grounding; their reasoning scores remain well below those of larger models.

## Sources

- [GitHub](https://github.com/firelex/jeff)

---
Canonical: https://techandbusiness.org/newswire/lO8WJSN60fPnamKC1WThP3
Published: 2026-09-29T01:09:08.778Z
Story chronology: 2026-09-28T20:23:36.000Z
Retrieved: 2026-09-29T03:08:10.954Z
Publisher: Tech & Business (techandbusiness.org)