# LLaDA2.0-Uni: A Unified MoE for Vision Tasks

> The new open-source model from inclusionAI uses a Mixture-of-Experts architecture to handle multiple vision tasks in a single, diffusion-based system.

Published by The Open Weights on Apr 22, 2026. Canonical: https://theopenweights.com/news/llada2-0-uni-4qj9

## Key facts

- Company: inclusionAI
- Model: LLaDA2.0-Uni
- Version: 2.0-Uni
- Category: Any-to-Any
- Modalities: Any-to-Any, Text → Image, Image Editing
- License: Apache 2.0 (Open weights, commercial use allowed)
- Context window: 8K tokens
- Architecture: LLaDA2MoeModelLM
- Precision: BF16
- Weights size: 32.6 GB
- Resolution: 1024×1024
- Significance: notable
- Published: 2026-04-22
- Last verified: 2026-06-17
- Hugging Face: https://huggingface.co/inclusionAI/LLaDA2.0-Uni
- Canonical URL: https://theopenweights.com/news/llada2-0-uni-4qj9

AI research group inclusionAI has released LLaDA2.0-Uni, a new open-source model aimed at unifying a range of visual AI tasks. Released under a permissive Apache 2.0 license, the model introduces a novel architecture for handling complex image-related operations within a single framework.

The core of LLaDA2.0-Uni is its use of a diffusion-based, Mixture of Experts (MoE) architecture. This design choice is significant because it allows the model to efficiently manage different tasks without needing to deploy separate, specialized models. Instead of chaining together distinct systems for understanding, creating, and modifying images, LLaDA2.0-Uni integrates these functions into one coherent system.

## A Unified Approach

The model's key feature is its versatility. LLaDA2.0-Uni is designed to perform three primary categories of visual tasks:

*   **Image Understanding:** Analyzing and interpreting the content of an image.
*   **Text-to-Image Generation:** Creating new images from textual descriptions.
*   **Image Editing:** Modifying existing images based on instructions.

This unified capability represents a step toward more consolidated and flexible multimodal systems. By combining these functions, developers can build more streamlined applications that require a mix of generative and analytical vision. The complete [model is available on Hugging Face](https://huggingface.co/inclusionAI/LLaDA2.0-Uni) for researchers and developers to explore.

## Get the model

- [Hugging Face](https://huggingface.co/inclusionAI/LLaDA2.0-Uni)

## Sources

- [inclusionAI/LLaDA2.0-Uni](https://huggingface.co/inclusionAI/LLaDA2.0-Uni) — Hugging Face, Apr 22, 2026

---
Source: The Open Weights (https://theopenweights.com/). Aggregated and written by Claude, curated by humans. Cite as: "LLaDA2.0-Uni: A Unified MoE for Vision Tasks", The Open Weights, Apr 22, 2026, https://theopenweights.com/news/llada2-0-uni-4qj9