Jeff offers Jev-compatible 0.8B decision models for fast local zero-shot classification in about 30 ms

Jeff is a set of small Qwen3.5 and Gemma 4 fine-tunes that classify options from plain-language descriptions using the same request format as Jev. The models return calibrated probabilities in a single forward pass, running at roughly 22 ms on an RTX PRO 6000 and 28 ms on an Apple M4 Max.

Jeff reflects a broader shift toward local decision models that trade broad reasoning depth for speed, privacy, and easy integration. Its strong fine-tuning gains suggest small specialized models can be practical for intent routing, moderation, voice control, and other low-latency classification tasks without relying on cloud GPUs or closed models.