# Models

> Codiv serves open System One models. Each has open weights and an open-source server, so you can always run it yourself.

Source: https://codiv.ai/docs/models

## OpenJev

OpenJev turns [DiffusionGemma 26B-A4B](https://huggingface.co/nvidia/diffusiongemma-26B-A4B-it-NVFP4), a discrete diffusion language model, into a System One model. It writes the answer template onto the model's canvas, leaves only the answer slots as noise and reads a probability distribution for every slot from a single denoising step. There is no generation and no parsing, so an answer can never fall outside your schema.

| | |
|---|---|
| Current version | `openjev-0.1` (public experiment) |
| Base model | DiffusionGemma 26B-A4B (mixture of experts, about 4B active parameters), NVFP4 |
| Context window | 65,536 tokens (state and questions together) |
| Question types | noul, choice (up to 128 options), score (2–10 levels) |
| License | Apache-2.0 for both the weights and the server |
| Source | [github.com/razorback16/openjev](https://github.com/razorback16/openjev) |

## Model ids

| Id | Serves |
|---|---|
| `openjev-latest` | Alias for the newest release. Currently `openjev-0.1`. |
| `openjev-0.1` | OpenJev 0.1, pinned. Use this if you want answers that don't change when a new version ships. |
| `jev-latest`, `jev-preview` | Aliases for `openjev-latest`, so the TypeSafe SDK defaults work unchanged. |

The response's `model` field always names the exact version that answered.

OpenJev 0.1 runs on an unmerged vLLM pull request ([#57250](https://github.com/vllm-project/vllm/pull/57250)). When that lands upstream we will release `openjev-1.0` and move `openjev-latest` to it; `openjev-0.1` stays available for pinning.

## Coming next

OpenJev 1.0, batch mode for large jobs, and more open System One models.