Open Jev An experimental inference harness for typed JSON decisions with DiffusionGemma. Denoise freely, then choose the most likely allowed tokens from the final logits. No training or fine-tuning. The goal is to explore how close an open diffusion model can get to TypeSafe AI's Jev on structured judgments. This is an independent experiment, not a reproduction of Jev's architecture, training, calibrated probabilities, or published workflow scores. Results at a glance Measured September 16, 2026 on one H100 80GB, BF16, using .
| Stars | 42 |
| Forks | 2 |
| Language | Python |
| Category | AI Tool |
| License | MIT |
| Quality Score | 60.7271168305411/100 |
| Open Issues | 1 |
| Last Updated | 2026-09-16 |
| Created | 2026-09-16 |
| Platforms | python |
| Est. Tokens | ~16k |
These tools work well together with open-jev for enhanced workflows:
Explore other popular ai tool tools:
open-jev is Typed JSON inference with DiffusionGemma, with Every and Jev benchmark results. It is categorized as a AI Tool with 42 GitHub stars.
open-jev is primarily written in Python.
You can find installation instructions and usage details in the open-jev GitHub repository at github.com/JoshuaSP/open-jev. The project has 42 stars and 2 forks, indicating an active community.
open-jev is released under the MIT license, making it free to use and modify according to the license terms.