Google DeepMind

Gemma 3 4B IT

Generally availableOpen weights131K context

Summary

Gemma 3 4B IT is an open-weight language model from Google DeepMind, released on 12 Mar 2025. It accepts text and images and generates text, with a 131K-token context window and up to 131K output tokens per response. As of 18 Nov 2026, the lowest listed API price is $0.05 per million input tokens and $0.10 per million output tokens (DeepInfra), across 3 providers we track.

Open Gemma instruction model for efficient chat and self-hosted deployments

Last verified 18 Nov 2026. Every figure on this page links to its source.

Specifications

DeveloperGoogle DeepMind
Release date12 Mar 2025 [models.dev]
StatusGenerally available [models.dev]
WeightsOpen weights [Hugging Face]
Licensegemma [Hugging Face]
Parameters4.3B [Hugging Face]
Context window131K (131,072 tokens) [models.dev]
Max output131K tokens [models.dev]
Knowledge cutoffAug 2024 [models.dev]
InputText and images [models.dev]
OutputText [models.dev]
Reasoning modeNo [models.dev]
Tool callingYes [models.dev]
Structured outputYes [models.dev]
Hugging Facegoogle/gemma-3-4b-it [Hugging Face]

Gemma 3 4B IT API pricing by provider

USD per million tokens. Sorted by input price.

ProviderInputOutputCached inputContextSource
DeepInfra
google/gemma-3-4b-it
$0.05$0.10–131Kmodels.dev, 18 Nov 2026
Hugging Face Inference Providers
google/gemma-3-4b-it
$0.05$0.10–131Kmodels.dev, 18 Nov 2026
OpenRouter
google/gemma-3-4b-it
$0.05$0.10–131KOpenRouter, 18 Nov 2026

Running Gemma 3 4B IT locally

Estimated memory, from the parameter count on Hugging Face. Includes about 0.8 GB of runtime overhead.

QuantizationWeightsTotal, 8K contextTotal, 32K context
Q4_K_M (4-bit)2.6 GB3.4 GB–
Q8_0 (8-bit)4.6 GB5.4 GB–
FP16 / BF16 (16-bit)8.6 GB9.4 GB–

At 4-bit with an 8K conversation, Gemma 3 4B IT needs about 3.4 GB, so it fits on a graphics card with 8 GB or more, such as the NVIDIA GeForce RTX 4060 8 GB. The model's architecture is not public, so the conversation cache is not included. Check your own hardware, or see Running local LLMs on an 8 GB GPU on our sister site AIPCs.

Frequently asked questions

What is the context window of Gemma 3 4B IT?

Gemma 3 4B IT has a context window of 131,072 tokens (131K) and can generate up to 131,072 tokens in a single response, according to models.dev.

How much does the Gemma 3 4B IT API cost?

Gemma 3 4B IT is listed from $0.05 per million input tokens and $0.10 per million output tokens (DeepInfra). Prices last verified 18 Nov 2026.

Is Gemma 3 4B IT open source?

Gemma 3 4B IT is an open-weight model: its weights are published on Hugging Face as google/gemma-3-4b-it under the gemma license. Check the license terms before commercial use.

When was Gemma 3 4B IT released?

Google DeepMind released Gemma 3 4B IT on 12 Mar 2025, according to models.dev.

Which providers offer Gemma 3 4B IT?

We track Gemma 3 4B IT on 3 providers: DeepInfra, Hugging Face Inference Providers, OpenRouter.

How much VRAM does Gemma 3 4B IT need?

About 3.4 GB at 4-bit (Q4_K_M) with an 8K-token conversation, and about 5.4 GB at 8-bit. The 4-bit model fits on a graphics card with 8 GB or more, such as the NVIDIA GeForce RTX 4060 8 GB. These are estimates from the model's published parameter count.

Similar models from other labs

Comparable on output type, weights, release date, price and context window. Prices are the lowest listed, USD per million tokens.

ModelLabReleasedContextInput $/1MOutput $/1MWeights
Command R7B ArabicCohere27 Feb 2025128K$0.04$0.15Open
DeepSeek R1 0528 Qwen3 8BDeepSeek29 May 2025128K$0.06$0.09Open
Command R7BCohere13 Dec 2024128K$0.04$0.15Open
Mistral Small 3.1 24BMistral AI17 Mar 2025128K$0.10$0.30Open
Baichuan M2 32BBaichuan13 Aug 2025131K$0.07$0.07Open
Llama-4-Scout-17B-16E-Instruct-FP8Meta5 Apr 2025128K$0.10$0.30Open

Change history

  • Tool calling changed from no to yes
  • Max output changed from 16K to 131K
  • Tool calling changed from yes to no
  • Max output changed from 131K to 16K
  • Max output changed from 16K to 131K
  • Tool calling changed from no to yes
  • Tool calling changed from yes to no
  • Max output changed from 131K to 16K
  • Tool calling changed from no to yes
  • Max output changed from 16K to 131K
  • Tool calling changed from yes to no
  • Max output changed from 131K to 16K
  • Max output changed from 16K to 131K
  • Tool calling changed from no to yes
  • Tool calling changed from yes to no
  • Max output changed from 131K to 16K
  • Max output changed from 16K to 131K
  • Tool calling changed from no to yes
  • OpenRouter price changed: input $0.04 → $0.05, output $0.05 → $0.10 per 1M tokens
  • Max output changed from 131K to 16K
  • Tool calling changed from yes to no
  • Max output changed from 118K to 131K
  • Tool calling changed from no to yes
  • No longer listed on nvidia
  • Tool calling changed from yes to no

Sources