$ cat wiki/models/diffusion-gemma.md
DiffusionGemma
modelupdated 2026-07-27created 2026-07-27
Google DeepMind's first open-weight text diffusion model, built on the Gemma 4 26B MoE backbone.
Spec
| Attribute | Value |
|---|---|
| Developer | Google DeepMind |
| Released | 2026-06-11 |
| Announced | 2026-06-11 |
| Context window | unknown |
| Pricing | free (open-weight) |
| License | Apache 2.0 |
| Availability | Hugging Face (google/diffusiongemma-26B-A4B-it), Kaggle, Google Cloud Vertex AI Model Garden |
Release Date
June 11, 2026.
Architecture
- Base: Gemma 4 26B-A4B Mixture-of-Experts (25.2B total params, ~3.8B active per forward pass)
- Decoding: Discrete text diffusion — generates 256-token blocks in parallel via iterative denoising (Uniform State Diffusion), not left-to-right autoregressive
- Memory: ~18 GB VRAM (MXFP4 quantization) — significantly lower local deployment barrier than comparable models
- Framework: native vLLM support
Benchmarks
DiffusionGemma trades accuracy for throughput vs. the autoregressive Gemma 4 26B-A4B baseline:
| Benchmark | DiffusionGemma | Gemma 4 26B-A4B (autoregressive) |
|---|---|---|
| AIME 2026 | 69.1% | 88.3% |
| MMMU Pro | 54.3% | 73.8% |
| Multi-step reasoning (math, vision-language) takes a larger accuracy hit. Pure text tasks suffer less. The tradeoff is the ~4× throughput gain. |
Performance
- 1,000+ tokens/sec on a single H100 GPU
- 700+ tokens/sec on RTX 5090
- ~4× faster than autoregressive Gemma 4 26B-A4B at matched benchmark quality for throughput-bound workloads
Use Cases
- High-throughput text generation workloads where latency or cost per token is the primary constraint
- Not suited for tasks requiring maximum reasoning accuracy (multi-step math, complex vision-language)
Compared To
| Model | Type | Speed (H100) | AIME 2026 | License |
|---|---|---|---|---|
| DiffusionGemma | Diffusion MoE | 1,000+ t/s | 69.1% | Apache 2.0 |
| Gemma 4 26B-A4B | Autoregressive MoE | ~250 t/s (est.) | 88.3% | Apache 2.0 |
| Kimi K3 | Autoregressive MoE | unknown | — | Modified MIT |
Significance
DiffusionGemma is the first open-weight text diffusion model from a major AI lab to enter production. Parallel block-decoding (256 tokens per iteration) challenges autoregressive as the only viable decode path at scale. Opens engineering trade-space: operators can now choose throughput over per-task reasoning quality, a tradeoff that didn't exist in open-weight production models before this release.
Sources
- Google DeepMind: DiffusionGemma model page
- Hugging Face model card: google/diffusiongemma-26B-A4B-it
- MarkTechPost: DiffusionGemma release
- The Register: text diffusion speed story
- Source snapshot: sources/blogs/deepmind-2026-06-11-diffusion-gemma.md