Google · language model
Gemma 3 4B: pricing, context and benchmarks
Gemma 3 4B is a language model from Google released on March 13, 2025. It costs $0.04 per million input tokens and $0.05 per million output tokens, with a 131.1K-token context window. Epoch AI ranks it #125 of 125 rated models.
| Performance (ECI) | 0/40 | |
|---|---|---|
| Price | 29/30 | |
| Context | 4/15 | |
| Features | 7/15 |
What you need to know about Gemma 3 4B
Gemma 3 4B was released by Google (United States) on March 13, 2025. It accepts text and images as input and produces text. Its context window reaches 131,072 tokens, larger than 29 % of the models in the registry, and it can write up to 117,964 tokens in one go.
Released more than 1 year ago, it belongs to a generation most vendors have since superseded.
Gemma 3 4B pricing
On OpenRouter, Gemma 3 4B costs $0.04 per million input tokens and $0.05 per million output tokens. Counting three tokens read for every token written, its average price is $0.042 per million, more expensive than 2 % of paid models (18.2 times cheaper than the median of $0.76).
A concrete example: a task that sends 2,000 tokens (about 1,500 words) and receives a 500-token answer costs $0.0001. At 1,000 tasks a day, budget $3 a month. The price calculator redoes this with your own volumes.
Tokens already sent and cached are billed at $0.02 per million.
Performance and benchmarks
The Epoch Capabilities Index (ECI), which aggregates published results across many benchmarks, gives Gemma 3 4B a score of 115.9 (range 97.4 to 123.6). That places it #125 of 125 rated models.
Scores published by Epoch AI on the benchmarks the site tracks:
| Benchmark | What it measures | Score | Date |
|---|---|---|---|
| GPQA Diamond | PhD-level science questions | 23.2 % | Mar 12, 2025 |
| OTIS Mock AIME | maths competition problems | 7.5 % | Mar 12, 2025 |
For at least equal performance, Mistral Nemo costs at least 30% less.
Features, privacy and local use
- It does not support tool calling: it is not suited to driving an AI agent that has to act.
- It can answer in an enforced JSON format (structured outputs).
- Its knowledge stops at August 2024.
Its weights are open and downloadable from Hugging Face: you can run it on your own machines without sending your data to a third party.
Other Google models
| Model | Released | Avg. price | ECI |
|---|---|---|---|
| Gemini 4 Argon | Oct 13, 2026 | $4 | 166.5 |
| Nano Banana 2.1 | Oct 6, 2026 | $3 | — |
| Gemini 3.8 Flash | Sep 2, 2026 | $1.65 | 157.4 |
| Gemini 3.7 Flash | Aug 13, 2026 | $1.5 | 158 |
| Gemini 3.6 Flash | Jul 21, 2026 | $1.5 | 154.5 |
| Gemini 3.5 Flash Lite | Jul 21, 2026 | $0.85 | 144.9 |
| Nano Banana 2 Lite (Gemini 3.1 Flash Lite Image) | Jun 30, 2026 | $0.56 | — |
| Nano Banana 2 (Gemini 3.1 Flash Image) | Jun 18, 2026 | $1.13 | — |
Verdict
Gemma 3 4B scores 40/100 on our grid, which only uses published figures.
Strengths
- Low price: cheaper than 98 % of paid models.
- Open weights: self-hostable, your data stays with you.
Limitations
- No tool calling: unusable as an agent's brain.
- Modest performance: behind 99 % of rated models.
- Older generation, often replaced by a newer model from the same vendor.
Avoid if…
- you are building an agent that must call tools.
Alternatives
- Mistral Nemo$0.03
- gpt-oss-20b$0.04
- Phi 4$0.09
- GPT-4o-mini$0.26
- GPT-4o-mini (2024-07-18)$0.26
- Qwen3 32B$0.13
Comparisons and tools
Learn more
Frequently asked questions
How much does Gemma 3 4B cost?
$0.04 per million input tokens and $0.05 per million output tokens on OpenRouter (checked October 25, 2026). A task with 2,000 tokens read and 500 written costs $0.0001.
How big is Gemma 3 4B's context window?
131,072 tokens, roughly 98,000 words, with a maximum of 117,964 output tokens.
Can Gemma 3 4B run locally?
Yes, its weights are published.
Is Gemma 3 4B suitable for an AI agent?
No: without tool calling it cannot act, only answer.
How good is Gemma 3 4B?
Epoch AI ranks it #125 of 125 rated models (ECI 115.9). On GPQA Diamond, it scores 23.2%.
What are the alternatives to Gemma 3 4B?
The closest models on price and performance are Mistral Nemo, gpt-oss-20b, and Phi 4.
Related articles
- Mistral AI: models, pricing and agentsOctober 25, 2026
- LLM benchmarks: how to read model rankingsOctober 24, 2026
- Running a local LLM: hardware, software, and limitationsOctober 24, 2026
- Open-source or closed models for your AI agents?October 5, 2026
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score