Version 3

Google DeepMind released DiffusionGemma on June 10, 2026 — an experimental open-weights 26B mixture-of-experts model that generates up to 256 tokens simultaneously by denoising from noise rather than predicting tokens one at a time. [2] The central claim is roughly 4x faster output than comparable autoregressive models, now confirmed by a third-party benchmark from atomic[.]chat on local H100 hardware in FP8 precision. [5] DeepMind explicitly frames the model as experimental and acknowledges output quality is lower than standard Gemma 4. [2] A wave of social media amplification has spread the release widely, but no independent quality evaluations have emerged.

DiffusionGemma is the most prominent open release to test whether diffusion-based text generation can deliver practically useful speed gains outside the lab. Third-party benchmarks are now confirming the speed claims, but the quality trade-off remains unquantified by any independent source — which is the central unknown for whether this architecture can move from developer curiosity to production use.

Google DeepMind released DiffusionGemma on June 10, 2026, as an experimental open-weights model built on the Gemma 4 26B mixture-of-experts architecture. Unlike autoregressive language models that predict one token at a time, DiffusionGemma generates text the way diffusion models generate images: it starts from a block of placeholder tokens and iteratively refines all 256 simultaneously. [1] Because the model activates only 3.8B of its 26B parameters per inference step, it fits within 18GB of VRAM when quantized, making it accessible on consumer hardware. [2][1]

The central performance claim is roughly 4x faster token output in single-user settings compared to equivalent autoregressive models. DeepMind reports over 1,000 tokens/sec on a single NVIDIA H100 and over 700 tokens/sec on a GeForce RTX 5090. [2] NVIDIA, which co-publicized the launch, explains the speed advantage in structural terms: pulling a full 256-token block through the transformer at once converts inference from a memory-bandwidth-bound workload into a compute-bound one. [3] Independent testing by Simon Willison via NVIDIA's free NIM cloud API yielded approximately 547 tokens/sec for a 2,409-token response. [4] A subsequent third-party benchmark by atomic[.]chat on local H100 hardware in FP8 precision also confirmed the 4x speed advantage against Gemma 4 26B A4B running autoregressively. [5]

DeepMind's own announcement is candid about limitations: output quality is lower than standard Gemma 4, and the model is explicitly not recommended for applications requiring maximum quality. [2] The intended use cases are developer experimentation, interactive local workflows, and tasks where bi-directional attention provides a structural edge — such as code infilling and in-line text editing. [2] NVIDIA's promotional coverage does not foreground these quality limitations. [3] Simon Willison identified DiffusionGemma as a public return of Gemini Diffusion research that Google briefly previewed in May 2025 but did not publicly follow up on until this release. [4]

The release has generated broad social media amplification across multiple languages in the days since launch, but this coverage has been largely amplification of the original speed claims rather than independent evaluation. No quality benchmarks or community evaluations have appeared, leaving the central trade-off — how much quality is sacrificed for the speed gain, and on which task types — unquantified.