diffusiongemma

About

What diffusiongemma is

Model Overview Description: DiffusionGemma 26B A4B IT is an open-weights multimodal generative model developed by Google DeepMind that processes text, image, and video inputs to produce text output via discrete diffusion. Built on the Gemma 4 26B A4B Mixture-of-Experts (MoE) architecture with 25.2B total parameters and 3.8B active parameters, the model employs an encoder-decoder design with bidirectional attention that generates tokens in parallel 256-token blocks, enabling high-speed generation exceeding 1,100 tokens per second at low batch sizes on NVIDIA Hopper H100 (FP8). DiffusionGemma 26B A4B IT supports a 256K token context window, configurable thinking (reasoning) mode, native function calling, and multilingual inference across 35+ languages. The NVIDIA DiffusionGemma 26B A4B IT NVFP4 model is quantized with Model Optimizer. This model is ready for commercial and non-commercial use. <br Third-Party Community Consideration This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party's requirements for this application and use case; see link to Non-NVIDIA DiffusionGemma 26B A4B Model Card. License/Terms of Use: SPDX-FileCopyrightText: Copyright (c) <year NVIDIA CORPORATION & AFFILIATES. All rights reserved. SPDX-License-Identifier: Apache-2.0 Governing Terms: Apache 2.0 Gemma Terms of Use and Gemma Prohibited Use Policy. Deployment Geography: Global Use Case: Use Case: DiffusionGemma 26B A4B IT is designed for developers, researchers, and enterprises requiring high-speed multimodal text generation. Supported use cases include conversational AI and chatbots, text summarization, code generation and step-by-step reasoning, image and document understanding (OCR, chart comprehension, PDF parsing, screen and UI parsing), video content analysis, agentic workflows with native function calling, and multilingual NLP tasks across 35+ languages. Release Date: Hugging Face 06/10/2026 via https://huggingface.co/nvidia/DiffusionGemma-26B-A4B-IT-NVFP4 Model Architecture: Architecture Type: Transformer Network Architecture: Mixture-of-Experts Total Parameters: 25.2B Active Parameters: 3.8B Vocabulary Size: 262,144 Base Model: Gem

Summarised from the published model card. Read the full card on the HuggingFace links below.

Specifications

What it is

How it works

How language models work

Variants

Sizes & precisions

Open weights ship in several sizes and precisions. One page, all the variants — pick the one that fits your GPU. VRAM figures are estimates from model size.

Using it via the API

Call it like any OpenAI endpoint

Once AxForge deploys nvidia-diffusiongemma for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (nvidia-diffusiongemma below is illustrative; you get the exact model name on deployment.)

$ curl -sS https://api.axforge.ai/v1/chat/completions \
  -H "Authorization: Bearer $AXFORGE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"nvidia-diffusiongemma","messages":[{"role":"user","content":"Hello"}]}'

Details

Languages, data & research

Licence

Open weights

Sources

Weights & code

Explore

More language models