Gemma 3 (4B) represents an advanced multimodal AI system that expertly combines capabilities in both text and visual comprehension. Equipped with a substantial 131,000-token context window, it is designed to handle extensive information processing and incorporates function calling to efficiently execute sophisticated tasks. Built for flexible deployment, Gemma 3 delivers a strong mix of high performance and resource efficiency, operating effectively across various platforms ranging from smartphones to high-end workstations.
Google prioritizes ethical AI development, emphasizing transparency about Gemma 3’s capabilities and limitations. Responsible usage is strongly encouraged to mitigate risks of misuse or harmful applications related to generated outputs.
Gemma 3 is distributed under the Gemma Terms of Use with a commercially-friendly license model. This license supports both research and commercial applications while ensuring adherence to established ethical standards.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "google/gemma-3-4b-it", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "google/gemma-3-4b-it", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"google/gemma-3-4b-it","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 4.8 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | September 22, 2026 |
| Coding | 2.7 | Composite score across standardised coding evaluations, measured independently by Artificial Analysis | Source | September 22, 2026 |
| Math | 12.7 | Composite score across standardised mathematics evaluations, measured independently by Artificial Analysis | Source | September 22, 2026 |
| MMLU | 74.5 | Multi-discipline knowledge across 57 school and professional subjects | Source | September 22, 2026 |
| MMLU-Pro | 45.3 | Multi-discipline knowledge + reasoning (harder MMLU) | Source | September 22, 2026 |
| MATH | 43.3 | Competition mathematics problems across five difficulty levels | Source | September 22, 2026 |
| GSM8K | 71.0 | Grade-school maths word problems requiring multi-step arithmetic | Source | September 22, 2026 |
| HumanEval | 45.7 | Python function synthesis from a docstring, scored by unit tests (pass@1) | Source | September 22, 2026 |
| DocVQA | 82.3 | Question answering over document images | Source | September 22, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Gemma 3 (4B) This page | Chat + assistants | |||
| Reasoning + agents | ||||
| Balanced coding + agents | ||||
| Long-context, multimodal & agentic workflows | ||||
| Reasoning + agents |
Gemma 3 (4B) has a 131,000 tokens context window and can return up to 32,768 tokens.
Gemma 3 (4B) takes image, text as input and returns text.
Use google/gemma-3-4b-it as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Gemma 3 (4B) became available on September 9, 2025.
Gemma 3 (4B) is priced at input $0.06877 / 1M tokens, output $0.13754 / 1M tokens, cached input $0.055016 / 1M tokens.
Yes, Gemma 3 (4B) can stream responses as they are generated.
Yes, Gemma 3 (4B) accepts image input alongside text.
Gemma 3 (4B) was built by Google.
Yes, it is designed for on-device deployment and efficient inference.
