Skip to main content
AI Choice Engine

Google

Gemma 4 31B

Gemma 4 31B is a Google model with 262K context, Not verified in / Not verified out per million tokens, last verified October 2, 2026. Open weights: yes. Public-board scores are linked to the publisher; vendor-only figures stay in the watch-out, not on the board.

Google’s instruction-tuned Gemma 4 31B open-weight variant for text/image input and video through frame sequences, generating text.

Recorded profile verification October 2, 2026Individual provider fields may change

AI model specification details
SpecificationGemma 4 31B
ProviderGoogle
TierBalanced
Context window262K
Max outputNot verifiedUnverified
Input / 1M tokensNot verifiedUnverified
Output / 1M tokensNot verifiedUnverified
WeightsOpen
Parameters30.7B total (dense)
Reasoning levelsNot verifiedUnverified
Modalitiestext, image, video
LicenseApache 2.0
ReleasedMarch 31, 2026

Pricing tiers: Downloadable instruction-tuned weights declare Apache 2.0 in the exact publisher card. Self-hosting needs compute; the family has no universal per-token hosting tariff. Hosted API identifiers, prices and generation caps depend on the route, and are not inferred from local examples.

Publisher evidence

Published benchmark results

Each result keeps its source date visible. Dates refer to source publication or capture unless the source explicitly states when it measured the result. A missing benchmark is not treated as a zero.

Published benchmark results for Gemma 4 31B
BenchmarkScoreSource dateSource
Artificial Analysis Intelligence IndexGemma 4 31B152026-10-01Artificial Analysis · View source

Best for

  • Self-hosted multimodal applications
  • Text generation
  • Instruction following

Watch out

Thinking is enabled by <|think|> in the system prompt; omit it to disable. Disabled thinking still emits an empty thought block. Use temperature=1.0/top_p=0.95/top_k=64. Omit historical thoughts except during tool-call turns. Video understanding processes frame sequences; this variant has no audio encoder or native audio support. Native text context comes from this checkpoint’s text_config, not the vision encoder’s positional limit. Exact hard output cap and universal hosting rates remain unknown; example generation budgets are not caps. Vendor benchmark runs remain unranked until identity/settings are reconciled.

Source receipts

Catalog figures for this model were checked against the following sources.

Common questions

Gemma 4 31B

Answered from the verified figures on this page rather than general guidance.

What is Gemma 4 31B's context window?
Gemma 4 31B accepts about 262K tokens of context. That only matters if you routinely send very long documents, large codebases, or multi-turn histories that approach that limit.
What is Gemma 4 31B best for?
Gemma 4 31B is a balanced tier from Google. It suits self-hosted multimodal applications, text generation, instruction following. Thinking is enabled by <|think|> in the system prompt; omit it to disable. Disabled thinking still emits an empty thought block. Use temperature=1.0/top_p=0.95/top_k=64. Omit historical thoughts except during tool-call turns. Video understanding processes frame sequences; this variant has no audio encoder or native audio support. Native text context comes from this checkpoint’s text_config, not the vision encoder’s positional limit. Exact hard output cap and universal hosting rates remain unknown; example generation budgets are not caps. Vendor benchmark runs remain unranked until identity/settings are reconciled.
Can I self-host Gemma 4 31B?
Gemma 4 31B publishes open weights, but self-hosting depends on the licence, hardware footprint, quantisation quality, and serving stack. A hosted API is often cheaper until you have measured throughput and concurrency on your own hardware.
Model comparison catalogBenchmark library