Chat with Gemma 4 31B.

Google's newest Gemma. Reads pictures as well as text, and can show its thinking.

Start chatting with Gemma 4 31B now — free, no account

At a glance

  • Maker: Google
  • Released: 2026
  • Parameters: 31B
  • Context window: 256k
  • Size: Medium
  • Memory: Handles long documents
  • Good at: Writing, and reading images
  • Supports: tools, images, zero data retention
  • License: Gemma Terms of Use (Google)
  • Upstream API price: $0.40 per million tokens

Announced in Gemma 4: Byte for byte, the most capable open models.

What is Gemma 4 31B?

Gemma 4 31B is Google's newest mid-size open model, and the dense flagship of the Gemma 4 family. Gemma is Google's open-weight line — close kin to its closed frontier models, but packaged for people who want to run, study and build on the weights themselves. Where the Gemini line chases the frontier, Gemma chases the sweet spot, and the 31B is where the family's writing quality and careful reading live.

As a dense model, all 31 billion parameters contribute to every token, which is part of why its prose has the well-proofread character Gemma is known for. The context window is 256k tokens — roughly 190,000 words, enough for a long novel or a serious code module — and it reads pictures as well as text. It can also show its thinking: the model is built to expose its reasoning, which UsingOpen surfaces in the answer panel. Tool support and zero-data-retention terms round out the picture, at $0.40 per million tokens on Google's API.

Within the Gemma 4 pair on UsingOpen, the 31B is the careful one and the 26B is the quick one — a mixture-of-experts design that wakes only 3.8B of its 25B parameters per token. Both cost the same $0.40, which makes the choice unusually clean: same price, different temperaments. Pick the 31B when the words matter; pick the 26B when the clock does.

What can you do with Gemma 4 31B?

The 31B's strengths cluster around text quality and careful reading:

  • Writing and editing: prose, documentation, email and marketing copy with a consistent voice.
  • Reading pictures — screenshots, diagrams, scanned pages — and discussing them precisely.
  • Long-form work inside a 256k-token window, roughly 190,000 words: whole books, long contracts, large documents.
  • Step-by-step problem solving with visible reasoning you can check rather than trust blindly.
  • Tool-using workflows that need a model which reads instructions literally and well.
  • Studying and explaining — the model's visible reasoning traces make it a decent tutor.

How does Gemma 4 31B compare?

The in-family comparison is the interesting one: Gemma 4 26B costs the same $0.40 per million tokens but trades density for speed, activating 3.8B of its 25B parameters per token and reading video as well as pictures. The 31B is the more careful writer and thinker; the 26B is the faster responder. Google pricing them identically is close to a public invitation to try both and see which temperament fits.

Across makers, the natural peers are other careful mid-size models: Alibaba's Qwen3.8 27B ($3.20) reasons harder per token but costs eight times as much, and Meta's Muse Glimmer 30B ($1.50) leans toward agents on modest hardware. For writing-heavy work with occasional images, the 31B is arguably the best value in this weight class on the site.

Is Gemma 4 31B right for you?

The 31B is the Gemma for people who care about words: documentation, editorial work, long emails, careful summaries, anything where the difference between decent prose and good prose is the deliverable. It is also a fine pick when you want reasoning you can audit — the visible thinking means you can watch it work — and its $0.40 price makes it one of the least expensive careful models on the site. Pictures are read natively, tools are supported, and the 256k window is honest about what most documents actually need.

Choose the 26B sibling instead when response time dominates, or a million-token model when archive-scale context matters more than prose polish. If your work is agent-first — tool loops on modest hardware — Meta's Muse Glimmer aims at that job more directly. But for writing-heavy work with occasional images, the 31B is the model this catalogue would hand you first, and the price is close to an apology.

Frequently asked questions

What is the context window of Gemma 4 31B?

256k tokens — roughly 190,000 words. That is a long novel, a lengthy contract, or a substantial code module with everything in view at once. For comparison, that is about a novel and a half — or a serious code module with its tests and docs — in one call, which covers most real documents without triage.

Does Gemma 4 31B support image input?

The underlying model supports image input as well as text. Note that image attachments are not yet enabled in UsingOpen's web chat. Scanned pages and whiteboard photos are the everyday wins — material that defeats text-only models at the door.

Is Gemma 4 31B a reasoning model?

It can show its thinking. The model is built to expose its step-by-step reasoning, and UsingOpen displays that reasoning alongside the answer, so you can check the work rather than take it on faith.

Who makes Gemma 4 31B?

Google, as part of Gemma — its family of open models built on the research behind its larger frontier systems. The Gemma line is aimed at developers and researchers who want capable weights they can run and inspect themselves.

How does Gemma 4 31B compare to Gemma 4 26B?

Both cost $0.40 per million tokens. The 31B is dense — every parameter active on every token — which gives it more careful writing and thinking. The 26B is a mixture-of-experts that activates only 3.8B of its 25B parameters per token, making it much faster. Care versus speed, at the same price. Same price, opposite temperaments is an unusually clean offer, and it means the right answer is usually to try both for a day rather than agonise.

What licence is Gemma 4 31B under?

Google's own Gemma Terms of Use — a custom, permissive-style licence rather than Apache or MIT. It allows most commercial and research use, with Google's conditions attached, and the weights are freely downloadable. The terms are permissive for the vast majority of uses, with conditions rather than restrictions, and the weights are freely downloadable.

Is Gemma 4 31B free to chat with on UsingOpen?

Yes. Pick it in the model picker and start chatting — no account, no payment. History is saved only on your device. At $0.40 upstream it is also one of the more sustainable free models — cheap enough that capacity pressure rarely bites.

Related models on UsingOpen

Gemma 4 31B sits in busy company — the closest alternatives on UsingOpen are:

  • Gemma 4 26B by Google: the mixture-of-experts Gemma — 25B total, 3.8B awake, quick on a budget.
  • Qwen3.8 27B by Alibaba: the newest mid-size Qwen — a dense 27B that thinks before it answers.
  • Muse Glimmer 30B by Meta: the compact agent from Meta Superintelligence Labs, distilled to run on ordinary hardware.

Head-to-head comparisons

See how Gemma 4 31B lines up with the models people most often weigh it against:

Model availability changes as upstream capacity comes and goes — the picker on the chat page shows what is ready right now.

Chat with Gemma 4 31B free on UsingOpen — no account, no sign-up

← All open-weight models · Compare models · Free AI chat · FAQ