# DiffusionGemma

15 Hacker News commenters recommended DiffusionGemma and 7 criticized it, November 2025 to November 2026. It was recommended in 5 of 13 months. It meets the Hacker Approved badge rule.

**Verdict:** Diffusion-based open model people recommend for very fast token output, especially on a single GPU. Common caveat: output quality often trails regular autoregressive Gemma models.

## Why people recommend it

- Very fast token generation (9 comments): "Instead, diffusion models work much faster for individual prompts, and not by a small margin either." ([source](https://news.ycombinator.com/item?id=48556697)) "diffusion is huge for local MoE inference because of the improved token generation efficiency" ([source](https://news.ycombinator.com/item?id=48490055)) "sadly isn't as smart as Gemma itself, but holy hell is it FAST" ([source](https://news.ycombinator.com/item?id=49066764))
- Fast on a single GPU (5 comments): "single-user scenarios, which is where these diffusion models really make a large difference in performance." ([source](https://news.ycombinator.com/item?id=48560805)) ">900TPS output reliably on a single 5090 with the nvfp4 model." ([source](https://news.ycombinator.com/item?id=49378908)) "it really makes good use of single GPUs for single responses" ([source](https://news.ycombinator.com/item?id=49375028))
- Good responses for its size (2 comments): "seems to perform better (in responses) than benchmarks seemed to indicate and reasonably fast for its model size" ([source](https://news.ycombinator.com/item?id=48853830)) "I like the model a lot and it's fairly good at reasoning." ([source](https://news.ycombinator.com/item?id=49376900))

## What people warn about

- Output quality trails autoregressive models (4 comments): "I'd still eat the time cost of the normal Gemma for 10% extra performance" ([source](https://news.ycombinator.com/item?id=48479890)) "performed a lot worse than its equivalent autoregressive counterpart" ([source](https://news.ycombinator.com/item?id=50156658)) "sadly isn't as smart as Gemma itself, but holy hell is it FAST" ([source](https://news.ycombinator.com/item?id=49066764))
- Large drops on hard benchmarks (2 comments): "Look at the dgemma benchmarks - pretty steep dropoffs and the more difficult the benchmark the worse the dropoff." ([source](https://news.ycombinator.com/item?id=48560517)) "Large dropoffs, with the hardest benchmarks seeing the largest drops." ([source](https://news.ycombinator.com/item?id=48479338))

## Alternatives hackers recommend

- [Inception Mercury 2](https://hackerapproved.com/p/mercury-2.md): 64 recommend
- [Inception Labs](https://hackerapproved.com/p/inception-labs.md): 5 recommend

## Cite as

Hacker Approved, "DiffusionGemma", https://hackerapproved.com/p/diffusiongemma/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).