Gemma 3 4B
Open weightVisionGoogle · Released March 2025 · Knowledge cutoff August 2024
Gemma 3 introduces multimodality, supporting vision-language input and text outputs. It handles context windows up to 128k tokens, understands over 140 languages, and offers improved math, reasoning, and chat capabilities,...
It is part of Google's Gemini and Gemma catalog: Pro tiers lead on capability, Flash tiers on speed and price, and Gemma models are open weight.
Facts and pricing
- Provider
- Context window
- 131K tokens
- Input price
- 0.55 credits / 1M tokens ($0.050 raw)
- Output price
- 1.1 credits / 1M tokens ($0.10 raw)
- Vision (image input)
- Yes
- Tool calling
- No
- Extended reasoning
- No
Credits are what Zeplik bills: 1 credit = $0.10, computed from the raw provider rate with a 1.10x margin. Raw prices shown per 1M tokens.
Try Gemma 3 4B now
What Gemma 3 4B is best for
- Everyday chat and drafting on an open-weight model with transparent lineage
- Working with images: screenshots, charts, photos and scanned documents alongside text
Example prompts
Prompts that suit a open weight model like Gemma 3 4B:
- Draft a project update from these rough notes
- Explain how HTTPS works to a curious teenager
- Turn this list of features into a changelog entry
- Brainstorm objections to this proposal and how to answer them
Google family
| Model | Context | Input cr/M | Output cr/M | Released |
|---|---|---|---|---|
| Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) | 66K | 2.8 | 16.5 | June 2026 |
| Nano Banana 2 (Gemini 3.1 Flash Image) | 131K | 5.5 | 33.0 | June 2026 |
| Nano Banana Pro (Gemini 3 Pro Image) | 66K | 22.0 | 132 | June 2026 |
| Gemini 3.5 Flash | 1.0M | 16.5 | 99.0 | May 2026 |
| Gemini 3.1 Flash Lite | 1.0M | 2.8 | 16.5 | May 2026 |
| Gemma 4 26B A4B (free) | 262K | 0 | 0 | April 2026 |
| Gemma 4 26B A4B | 262K | 0.66 | 3.6 | April 2026 |
| Gemma 4 31B (free) | 262K | 0 | 0 | April 2026 |
| Gemma 4 31B | 262K | 1.3 | 3.8 | April 2026 |
| Lyria 3 Pro Preview | 1.0M | 0 | 0 | March 2026 |
Frequently asked questions
- What is Gemma 3 4B?
- Gemma 3 4B is an AI model by Google. Gemma 3 introduces multimodality, supporting vision-language input and text outputs. It handles context windows up to 128k tokens, understands over 140 languages, and offers improved math, reasoning, and chat capabilities,...
- How much does Gemma 3 4B cost on Zeplik?
- Input tokens cost 0.55 credits per million and output tokens 1.1 credits per million (1 credit = $0.10; the raw provider rates are $0.050 and $0.10 per million). New accounts start with free credits.
- How long can a conversation with Gemma 3 4B be?
- Gemma 3 4B has a 131K-token context window (131,072 tokens), which covers the conversation plus any documents you attach.
- Does Gemma 3 4B support images and tools?
- Gemma 3 4B accepts image input and does not support tool calling.
Related models
Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) is Google's fastest, most cost-efficient Gemini image model, built for high-velocity developer pipelines and rapid-fire visual exploration. It delivers text-to-image generation...
Gemini 3.1 Flash Image, a.k.a. "Nano Banana 2," is Google’s latest state of the art image generation and editing model, delivering Pro-level visual quality at Flash speed. It combines advanced...
Nano Banana Pro is Google’s most advanced image-generation and editing model, built on Gemini 3 Pro. It extends the original Nano Banana with significantly improved multimodal reasoning, real-world grounding, and...
Gemini 3.5 Flash is Google's high-efficiency multimodal model, bringing near-Pro level coding and reasoning at Flash-tier cost and speed. It is highly optimized for coding proficiency and parallel agentic execution...
Gemini 3.1 Flash Lite is Google’s GA high-efficiency multimodal model optimized for low-latency, high-volume workloads. It supports text, image, video, audio, and PDF inputs, and is designed for lightweight agentic...
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...