firelex / jeff

Jeff 
 Millisecond decisions. Any domain. 
 jeffhub.ai ·
 Hugging Face ·
 Jeff-Code ·
 Adapters ·
 GGUF ·
 Quick start ·
 Changelog 
 New: Jeff-Code , Jeff inside a coding agent with Qwen 3.8-27B:
 47% faster (32% less time) per task, same pass rate (62.4% vs 62.8%, 1,242 tasks, 6 benchmarks).
 Results → 
> **Jeff v1.3 (changelog).** A new adapter-first base, 15 adapters, GGUF files for llama.cpp, and
> Jeff-Code. v1.3 adapters work only on the v1.3 base; the v1.2 models stay available under their old names.
## Put Jeff in front of your 27B
Jeff is a 0.8B open "System 1" model. Let a strong local model such as Qwen3.8-27B do the writing and planning, and let
Jeff make the quick decisions in front of it. Across 8 adapters\*, with the same test rows each way:
| | Qwen3.8-27B makes every decision | **Jeff + adapter alone** | Jeff first, the 27B only when Jeff is unsure |
|---|---:|---:|---:|
| **Accuracy** (mean of 8 adapters\*) | 86.6% | **94.6%** | 95.0% |
| **Time per decision** (mean, Apple M4 Max) | 8.1 s | **0.25 s: 35× faster** | 0.46 s: 28× faster |
| **Memory** | 28.6 GB | **+2.09 GB** for Jeff with all 15 adapters loaded (1.83 GB with 3)\*\* | |
Jeff alone already beats the 27B on every task but one; with the 27B as a fallback, Jeff passes on only the questions
it is unsure about.
| Task | 27B alone | Jeff + adapter | Jeff, 27B as fallback | Passed on |
|---|---:|---:|---:|---:|
| **guard:** prompt injection and jailbreaks | 84.0% 3.9 s | **98.7%** 0.07 s | 98.7% 0.07 s | 0% |
| **triage:** urgency and sentiment | 81.3% 3.6 s | **88.7%** 0.11 s | 88.7% 0.11 s | 0% |
| **support-intents:** what the customer wants | 86.0% 6.4 s | **94.3%** 0.20 s | 94.3% 0.20 s | 0% |
| **tools:** which tool an agent should call | 90.3% 11.3 s | **98.3%** 0.32 s | 98.3% 0.32 s | 0% |
| **ground:** is the answer supported by the sources? | 96.7% 13.2 s | 95.0% 0.43 s | **98.0%** 2.09 s | 11% |
| **nav:** voice commands to on-screen items | 91.3% 7.2 s | **98.7%** 0.23 s | 98.7% 0.23 s | 0% |
| **spam:** spam and phishing | 88.0% 2.7 s | **98.7%** 0.08 s | 98.7% 0.08 s | 0% |
| **legal-clauses:** contract clause types | 75.0% 16.5 s | **84.8%** 0.58 s | 84.8% 0.58 s | 0% |
| emotion\*: the strongest of 27 emotions, or neutral | 35.6% 4.8 s | **59.6%** 0.14 s | 59.6% 0.14 s | 0% |
\*\*Jeff's memory is GPU memory after loading, measured on an RTX PRO 6000 with the released adapter files; the base
alone takes 1.74 GB.
\*Emotion is left out of the averages: picking the single strongest of 27 emotions (or neutral) in short Reddit
comments is hard even for people, and the human labels often disagree. Including it makes the gain larger, not smaller.
**How this was measured.**
- **Setup:** an Apple M4 Max with 128 GB, both models on MLX, one at a time. Qwen3.8-27B in 8-bit, prompted to answer
 directly (step-by-step reasoning off). These are Mac times; on an NVIDIA GPU, Jeff takes tens of milliseconds per
 decision (see Speed and size).
- **Sample:** each task uses a