Google model product
DiffusionGemma 26B-A4B is Google DeepMind's experimental open-weights text diffusion model based on the Gemma 4 26B-A4B Mixture-of-Experts architecture. It uses discrete diffusion to denoise 256-token canvases in parallel, targeting low-latency local and low-concurrency generation workloads with up to 4x faster text generation on dedicated GPUs. The model has 25.2 billion total parameters, 3.8 billion active parameters, a 256K context window, and multimodal text and image inputs.
Updated Aug 10, 2026. Default version: DiffusionGemma 26B-A4B
Structured fields from the published default version.
This profile uses the latest version under this unique model that has a calculated LLMBoard score. Arena and price are excluded.
Published benchmark records for the scored version DiffusionGemma 26B-A4B.
| MRCR v2 | 0.3 | 2 | 3 | 50.0% | C | |
| BIG-Bench Extra Hard | 0.5 | 4 | 11 | 70.0% | C | |
| MedXpertQA | 0.5 | 7 | 12 | 45.5% | C | |
| OmniDocBench 1.5 | 0.3 | 13 | 16 | 20.0% | C | |
| AIME 2026 | 0.7 | 15 | 17 | 12.5% | C | |
| CodeForces | 0.5 | 16 | 16 | 0.0% | C | |
| MathVision | 0.7 | 16 | 32 | 51.6% | C | |
| t2-bench | 0.6 | 20 | 23 | 13.6% | C | |
| LiveCodeBench v6 | 0.7 | 31 | 53 | 42.3% | C | |
| MMMLU | 0.8 | 37 | 49 | 25.0% | C | |
| MMMU-Pro | 0.5 | 54 | 65 | 17.2% | C | |
| MMLU-Pro | 0.8 | 63 | 129 | 51.6% | C | |
| Humanity's Last Exam | 0.1 | 77 | 92 | 16.5% | C | |
| GPQA | 0.7 | 117 | 233 | 50.0% | C |
Preference and agent-evaluation signals from published Arena datasets.
The default version has no published Arena rows, or its source alias has not been resolved.
Official vendor API PAYG pricing is summarized first. The table then lists individual provider offerings without treating their minimum as the official price.
The default version has no provider offering with current input or output token prices.
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
All published versions linked to this unique model. The score columns identify the version used by the current overall ranking.
| DiffusionGemma 26B-A4B | 38.3 | 25.2B | N/A | N/A | No | Apache 2.0 |
A concise description based on the published model registry.
DiffusionGemma 26B-A4B is Google DeepMind's experimental open-weights text diffusion model based on the Gemma 4 26B-A4B Mixture-of-Experts architecture. It uses discrete diffusion to denoise 256-token canvases in parallel, targeting low-latency local and low-concurrency generation workloads with up to 4x faster text generation on dedicated GPUs. The model has 25.2 billion total parameters, 3.8 billion active parameters, a 256K context window, and multimodal text and image inputs.
Use the benchmark, Arena and pricing sections above as separate evidence. A missing field means the current data snapshot does not support that claim.
Data snapshot: 2026-08-07. Editorial model content is not available in the backend.
Open a comparison with the three ranked models immediately above and below this model.
Recommendations prioritize the same model type and family, then the closest published LLMBoard score.
Common questions about DiffusionGemma 26B A4B.
DiffusionGemma 26B A4B's default version was released on Jun 10, 2026.
No official standard PAYG price is currently available for DiffusionGemma 26B A4B.
DiffusionGemma 26B A4B is published under Google in the model registry.
The current registry does not publish a context window for the default version.
No. The default version is not marked as having publicly available weights.
No published provider offering is currently linked to the default version.
Nearby ranked alternatives include Qwen3 30B A3B, Gemma 3 12B, Jamba 1.5 Large.