# Maisa: djev

> A typed-decision layer for Google's DiffusionGemma, from David Villalón at Maisa AI. It compiles a request into a small answer canvas, runs one denoising read on patched vLLM and reads the probabilities of the allowed labels, for text, images and images offered as options.

- Page: https://systemonemodels.tech/maisa/djev
- API: https://api.systemonemodels.tech/v1/models/maisa/djev
- Download: `pip install systemonemodels && systemone pull maisa/djev`

## Facts

| | |
|---|---|
| Maker | Maisa (https://systemonemodels.tech/maisa) |
| Decides | choice, score, noul |
| Architecture | djev |
| Base model | google/diffusiongemma-26b-a4b-it |
| Parameters | 26B |
| Licence | apache-2.0 |
| Availability | Open weights + hosted API |
| Released | 2026-09-19 |
| Latest version | 2026.09 |

## Hosted API

- Provider: Maisa (https://djev.dev)
- Docs: https://api.djev.dev/docs
- Input: $0.035/MTok · Output: Free

## Model card

<!-- generated by scripts/seed_catalog.py; edit content/models/catalog.yaml -->

# djev

A typed-decision layer for Google's DiffusionGemma, from David Villalón at Maisa AI. It compiles a request into a small answer canvas, runs one denoising read on patched vLLM and reads the probabilities of the allowed labels, for text, images and images offered as options.

djev adds no trained weights: it runs Google's diffusiongemma-26B-A4B-it (Apache-2.0) with a typed request compiler, runtime fixes and a local playground; reference hardware is one B200 in BF16. The author publishes no benchmark of this release and says an earlier sub-100 ms measurement used a different, quantized configuration. The hosted API at djev.dev is an invite-only research preview at $0.035 per million input tokens, output free. JevBench v1.4.2 ranks djev tenth (84.0% public, 29.9% sealed) and notes the docs call its probabilities experimental and uncalibrated.

## What it decides

- **choice** — picks one option from a set
- **score** — places the input on an ordered scale
- **noul** — answers a yes/no question with one calibrated probability

## At a glance

| | |
|---|---|
| Parameters | 26B |
| Base model | `google/diffusiongemma-26b-a4b-it` |
| Maker | Maisa |
| Released | 2026-09-19 |
| License | apache-2.0 |

## Hosted API

Served by **Maisa** — $0.035/MTok input, $0/MTok output. [Get access](https://djev.dev) · [API docs](https://api.djev.dev/docs).

## Read more

- [Code](https://github.com/Davipar/djev-dev)
- [Hosted preview](https://djev.dev)

---

*This page was opened by System One for Maisa, who can claim the organisation and take it over at any time.*

---

From System One Models — https://systemonemodels.tech/ · every System One model: https://systemonemodels.tech/system-one-models