Gemma models

The family

Gemma is Google's open-weight line, built from the research behind Gemini and published as downloadable checkpoints under the Gemma terms. Where Qwen ships a model for every modality, Gemma ships fewer and larger dense models, and puts the effort into what a single checkpoint does well.

One Gemma model is on offer here. It is the multimodal 31B instruct checkpoint, served from a published FP8 quantization of the original weights rather than one we made.

What we serve

Size and shape

30.7B dense parameters - every one of them active on every token, unlike a mixture-of-experts model of nominally similar size. It is the largest dense checkpoint we serve, and the FP8 build fits a 48 GB-class card.

Input

Text and images. There is no audio path and no video path in these weights, so a row carrying either has to go to a Qwen model instead - see the Qwen page.

Images cost a fixed amount

This is the practical reason to choose it. Gemma encodes every image to exactly 280 tokens, whatever its size, instead of tiling a large picture into many more. A page scan and a thumbnail bill the same, which makes an image-heavy job's estimate almost entirely a function of how many images it carries rather than what they happen to be.

Output

Text, with a reasoning trace that a job can cap with a reasoning budget the same way the Qwen reasoning models can.

Benchmark scores

Published figures, each from the source named beside it. Different benchmarks and different harnesses, so read down the column rather than across it.

The shape of that table is the point: strong on maths, knowledge and assistant-style work, well behind the 27B Qwen checkpoints on agentic coding. Artificial Analysis also measured it as unusually terse for its score - it reached index 39 on 39M output tokens where a higher-scoring sub-32B peer needed 98M. On a batch bill, where output tokens are the expensive half, a model that says less for the same answer is worth more than its index suggests.

When to pick it

Image-heavy jobs with a budget to hold

The fixed 280-token image cost makes the estimate predictable in a way tiling encoders are not, especially when the images vary wildly in size.

Maths, science and assistant-style answers

This is where it leads the models we serve at its size.

Long, repetitive runs where verbosity is the bill

Terse output at a competitive score is exactly the trade a batch queue wants.

Not for

Audio, video, or code and tool-use heavy work. Send those to Qwen, or to GLM for the hardest reasoning.

Pass gemma4-31b exactly as written when you submit a job. Its rates and its minimum job price are on the pricing page.