SAVRN
Search Contact SAVRN

Open-weight model · Image and text to text

gemma-4-31B-it

by Google google/gemma-4-31B-it

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output.

Parameters31.3B
Context262,144
Weights62.5 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads9M

Runs On

What it takes to serve gemma-4-31B-it (31.3B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

PrecisionWeightsMemory neededCheapest setupPer hourAlso fits
16-bit 62.5 GB 75.1 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 31.3 GB 37.5 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 15.6 GB 18.8 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

SAVRN's Notes on gemma-4-31B-it

One MI300X at $1.85 per hour carries this 31.3 billion parameter Gemma at 16-bit: 62.5 GB of weights and 75.1 GB needed against the card's 192 GB. Quantizing brings that to 37.5 GB at 8-bit and 18.8 GB at 4-bit, but here it buys concurrent context, not fit. With 262,144 tokens of context and image input, we would stay at 16-bit and give the spare memory to long documents.

Apache 2.0 permits commercial use, modification and redistribution, provided the license and NOTICE file stay attached and significant changes are stated. It is the instruction-tuned build of google/gemma-4-31B, described at arXiv:2607.02770; audio input belongs to the E2B, E4B and 12B sizes, not this one. DeepInfra hosts it at $0.13 in and $0.38 out per million tokens, Novita at $0.14 and $0.40; the output rate is where your own facility makes its case.

SAVRN Research, 2026-09-18

Model Card

By Google, published under apache-2.0, revision 842da3794eaa.

Hugging Face | GitHub | Launch Blog | Documentation| Technical Report
License: Apache 2.0 | Authors: Google DeepMind

Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages.

Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: E2B, E4B, 12B, 26B A4B, and 31B. Their diverse sizes make them deployable in environments ranging from high-end phones to laptops and servers, democratizing access to state-of-the-art AI.

Gemma 4 introduces key capability and architectural advancements:

Read the full model card (3,425 words)

Configuration

Architecture
Gemma4ForConditionalGeneration
Context length (tokens)
262,144
Layers
60
Hidden size
5,376
Feed-forward size
21,504
Attention heads
32
Key/value heads
16
Head dimension
256
Vocabulary size
262,144
Sliding window (tokens)
1,024
Model type
gemma4

Identity and Version

Repository
google/gemma-4-31B-it
Publisher
Google
Task
Image and text to text
Modality
Image and text
Library
transformers
Parameters
31.3B parameters
Languages
Not stated by the source
Revision
842da3794eaa0b77d5f08bae87a17459d91ff475
First published
2026-03-11
Last updated
2026-07-20

Files and Weights

12 files, 62.6 GB in total. The weights are 2 files totalling 62.5 GB in safetensors.

Weights2 files · 62.5 GB
Configuration5 files · 126.9 KB
Tokenizer2 files · 32.2 MB
Documentation1 file · 28.0 KB
Other1 file · 18.7 KB
Repository1 file · 1.7 KB
Every file
FileTypeSizeSHA-256
model-00001-of-00002.safetensorsWeights49.8 GB eeef8791537b
model-00002-of-00002.safetensorsWeights12.8 GB 018912220f55
.eval_results/mmmu_pro.yamlConfiguration182 B —
config.jsonConfiguration4.6 KB —
generation_config.jsonConfiguration208 B —
model.safetensors.index.jsonConfiguration120.2 KB —
processor_config.jsonConfiguration1.7 KB —
README.mdDocumentation28.0 KB —
chat_template.jinjaOther18.7 KB —
.gitattributesRepository1.7 KB —
tokenizer.jsonTokenizer32.2 MB cc8d3a0ce364
tokenizer_config.jsonTokenizer3.1 KB —

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
62.5 GB
Download from Google

Released by Google through Kaggle. Read the license.

Built From

Evaluations

Each result is shown as reported, with the conditions its reporter stated. None is a SAVRN measurement. A comparison lines two results up only when their configuration, unit and setup are all stated and identical.

BenchmarkConditionsResultReported byRevisionDate
Idavidrein/gpqa Task diamondMetric diamondComparison conditions not established 84.3 Model Card
Reported by a third party		 Evaluated revision not stated 2026-04-02
LiquidAI/ifstruct-v1.0 Task ifstruct_v1Metric ifstruct_v1Comparison conditions not established 95.9 Liquid AI — IFStruct v1.0 blog (gemma-4-31B-it)
Reported by a third party		 Evaluated revision not stated 2026-06-30
MMMU/MMMU_Pro Task mmmu_pro_visionMetric mmmu_pro_visionComparison conditions not established 76.9 Model Card
Reported by a third party		 Evaluated revision not stated 2026-05-12
TIGER-Lab/MMLU-Pro Task mmlu_proMetric mmlu_proComparison conditions not established 85.2 Model Card
Reported by a third party		 Evaluated revision not stated 2026-04-02
cais/hle Task hleMetric hleSetup With searchComparison conditions not established 26.5 Model Card
Reported by a third party		 Evaluated revision not stated 2026-04-02
joelniklaus/LEXam-hard Task lexam_hardMetric lexam_hardSetup lighteval, LEXam paper prompts, one response per question, no tools; DeepSeek-R1-0528 judge; mean of the German and English means over the 518 questions, 0-100Comparison conditions not established 37.41 SwissLegalEvals per-sample details (lighteval)
Reported by a third party		 Evaluated revision not stated 2026-06-11
llamaindex/ParseBench Task chartMetric chartSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 15 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
llamaindex/ParseBench Task layoutMetric layoutSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 57.4 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
llamaindex/ParseBench Task meanMetric meanSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 62.4 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
llamaindex/ParseBench Task tableMetric tableSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 80.6 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
llamaindex/ParseBench Task text_contentMetric text_contentSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 89.9 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
llamaindex/ParseBench Task text_formattingMetric text_formattingSetup Pipeline name: gemma4_31b_vllm_with_layoutComparison conditions not established 69.3 ParseBench
Reported by a third party		 Evaluated revision not stated 2026-04-17
sapbot/ask-my-agent-bench-2 Task fallbackMetric fallbackComparison conditions not established 72.1 Not named
Reported by a third party		 Evaluated revision not stated 2026-06-23

Memory Requirements

PrecisionWeights in memory
As published62.5 GB
16-bit62.5 GB
8-bit31.3 GB
4-bit15.6 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Hosted Prices

HostInput / outputUnitObserved
DeepInfra$0.13 / $0.38input / output, per million tokensSep 18, 2026
Novita$0.14 / $0.40input / output, per million tokensSep 18, 2026

From the SAVRN Index.

Built on This Model

Compare gemma-4-31B-it

Questions About gemma-4-31B-it

How much GPU memory does gemma-4-31B-it need?

About 75.1 GB at 16-bit and 18.8 GB at 4-bit: the weights (31.3B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run gemma-4-31B-it on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use gemma-4-31B-it commercially?

Yes. gemma-4-31B-it is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

What is gemma-4-31B-it's context length?

262,144 tokens, from the maximum position embeddings in its published configuration.

Similar Models

Model · Image and text to text

Qwen3.6-27B-AWQ-INT4

Cyankiwi

Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience. This release delivers substantial upgrades, particularly in For more details, please refer to our blog post Qwen3.6-27B. Empty cells (--) indicate scores not yet available or not applicable. For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API. Qwen3.6 can be served via APIs with popular inference frameworks. In…

Open weights apache-2.0 29.3B parameters 262,144 tokens transformers
View model

Model · Image and text to text

Qwen2.5-VL-32B-Instruct-AWQ

Qwen

In addition to the original formula, we have further enhanced Qwen2.5-VL-32B's mathematical and problem-solving abilities through reinforcement learning. This has also significantly improved the model's subjective user experience, with response styles adjusted to better align with human preferences. Particularly for objective queries such as mathematics, logical reasoning, and knowledge-based Q&A, the level of detail in responses and the clarity of formatting have been noticeably enhanced. In the past five months since Qwen2-VL’s release, numerous developers have built new models on the Qwen2-VL vision-language models, providing us with valuable feedback. During this period, we focused on…

Open weights apache-2.0 33.5B parameters 128,000 tokens transformers
View model

Model · Image and text to text

Qwen2.5-VL-32B-Instruct

Qwen

In addition to the original formula, we have further enhanced Qwen2.5-VL-32B's mathematical and problem-solving abilities through reinforcement learning. This has also significantly improved the model's subjective user experience, with response styles adjusted to better align with human preferences. Particularly for objective queries such as mathematics, logical reasoning, and knowledge-based Q&A, the level of detail in responses and the clarity of formatting have been noticeably enhanced. In the past five months since Qwen2-VL’s release, numerous developers have built new models on the Qwen2-VL vision-language models, providing us with valuable feedback. During this period, we focused on…

Open weights apache-2.0 33.5B parameters 128,000 tokens transformers
View model

Model · Image and text to text

Qwen3.6-27B-FP8

Qwen

Following the February release of the Qwen3.5 series, we're pleased to share the first open-weight variant of Qwen3.6. Built on direct feedback from the community, Qwen3.6 prioritizes stability and real-world utility, offering developers a more intuitive, responsive, and genuinely productive coding experience. This release delivers substantial upgrades, particularly in For more details, please refer to our blog post Qwen3.6-27B. Empty cells (--) indicate scores not yet available or not applicable. For streamlined integration, we recommend using Qwen3.6 via APIs. Below is a guide to use Qwen3.6 via OpenAI-compatible API. Qwen3.6 can be served via APIs with popular inference frameworks. In…

Open weights apache-2.0 27.8B parameters 262,144 tokens transformers
View model

Model · Image and text to text

Qwen3.8-27B

Qwen

Following the widespread community adoption of the Qwen3.5 and Qwen3.6 series, we are pleased to introduce Qwen3.8, the most capable generation in the Qwen open-model family to date. Built on the architectural foundation of Qwen3.5, Qwen3.8 delivers substantial gains across coding, professional work, research, and long-horizon agentic tasks. Qwen3.8-27B brings these advances to a compact, deployment-friendly dense model: a native vision-language model that understands images and videos, with flexible thinking control, designed to carry complex, multi-step tasks through to completion with greater reliability. For streamlined integration, we recommend using Qwen3.8 via APIs. Qwen3.8 can be…

Open weights apache-2.0 27.8B parameters 262,144 tokens transformers
View model