Models · Google

Gemma 4 26B, run in Germany by AKI.IO.

Gemma 4 26B is an open-weight model by Google that reads text and images, released under the Apache-2.0 licence. On Evanesca, AKI.IO runs it in Germany, in German data centres according to AKI.IO, and Evanesca keeps no copy of your prompts.

Checked on 20 Oct 2026

Gemma 4 26B

by Google · Apache-2.0

Germany (AKI.IO) Checks 6/6

Input / M: €0.168
Output / M: €0.84
Quick question: €0.0004

Prices per million tokens, VAT included. Quick question: 300 tokens in, 400 out.

The model at a glance

Maker: Google
Licence: Apache-2.0
Size: 26B (4B active) parameters
Context: 256k tokens as the operator announces it
Type: Text + images
Tool calling: Yes measured by a real call
Reasoning: —
Run by: AKI.IO AKI.IO GmbH, Berlin — a German company; its shareholders are not published
Where it runs: Germany (AKI.IO)
Checks: 6/6 All six: Licence, Provenance, Security, GDPR, Benchmarks and Live test
Price: €0.168 per million input tokens · €0.84 per million output tokens VAT included; reasoning tokens count as output
Open weights: Yes published by the maker: an operator can run the model
Model card: huggingface.co/google/gemma-4-26B-A4B-it

Where your prompts go

Here is the route of a message sent to Gemma 4 26B, end to end. No call to the model leaves Europe: not as a backup, not when servers are busy.

From your browser to EvanescaEvanesca passes your message straight on to AKI.IO. It stores neither the prompt nor the answer: both stay in memory only while the answer is written.
AKI.IO, GermanyAKI.IO runs Gemma 4 26B in Germany: AKI.IO says inference runs on GPU infrastructure in German data centres (AKI.IO, read on 13 Oct 2026). AKI.IO GmbH is a German company in Berlin; its shareholders are not published. As a company under German law, it is out of the CLOUD Act's direct reach. AKI.IO states that “prompts, inputs, and generated outputs are processed in volatile memory and are not logged, stored, or analyzed by AKI.IO” (AKI.IO); its privacy policy keeps usage data — request ID, model, status, timestamps, token usage — for 90 days, for billing and service maintenance (AKI.IO's privacy policy).
Back to you, with its passportThe answer comes back with its passport: model, operator, country, and what was kept. Your history lives in your browser; about this message, Evanesca keeps only billing data — tokens and cost.

What it costs

€0.168 per million input tokens and €0.84 per million output tokens, VAT included. Three typical exchanges on Gemma 4 26B:

COST OF AN EXCHANGE · VAT INCLUDED
Example	Tokens in	Tokens out	Cost
A quick question	300	400	€0.0004
A long document to summarise	15,000	800	€0.0032
A message late in a long conversation	40,000	2,000	€0.0084

€2 of credit covers about 5,100 quick questions on this model. Reasoning tokens count as output, so a model that thinks first can cost more than these examples. The exact cost of each answer shows under it. All prices.

Licence

Google publishes Gemma 4 26B under the Apache-2.0 licence. Apache-2.0 is a permissive open-source licence: it allows commercial use.

The official model card is the reference.

Similar models

Same maker first, then the same type.

EU region of Mistral AI

Mistral Large 4

Mistral AI · Text + images · checks 4/6

€2.5133 · €7.7246 / M tokens

Germany (AKI.IO)

DeepSeek V4.1 Flash

DeepSeek · Text + images · checks 6/6

€1.008 · €3.024 / M tokens

Paris, France

Qwen3.8 27B

Qwen · Text + images · checks 6/6

€1.008 · €5.544 / M tokens

Switzerland (Infomaniak)

Apertus v1.5 70B

Swiss AI Initiative · Text + images · checks 4/6

€1.2936 · €4.62 / M tokens

Paris, France

Qwen3.6 35B

Qwen · Text + images · checks 6/6

€0.42 · €2.52 / M tokens

EU region of Mistral AI

Mistral Small 4

Mistral AI · Text + images · checks 5/6

€0.2772 · €1.1088 / M tokens

Every model, with its operator and where it runs

Frequently asked questions

Where does Gemma 4 26B run on Evanesca?

In Germany, at AKI.IO: AKI.IO says inference runs on GPU infrastructure in German data centres (AKI.IO, read on 13 Oct 2026). AKI.IO GmbH is a German company in Berlin, so your prompts stay out of the CLOUD Act's direct reach. Its shareholders are not published. No call to the model leaves Europe: not as a backup, not when servers are busy.

Does Evanesca keep what I send to Gemma 4 26B?

No. The prompt and the answer stay in memory only while the answer is written, and your history lives in your browser; about your messages, Evanesca keeps only billing data — tokens and cost. AKI.IO states that “prompts, inputs, and generated outputs are processed in volatile memory and are not logged, stored, or analyzed by AKI.IO” (AKI.IO); its privacy policy keeps usage data — request ID, model, status, timestamps, token usage — for 90 days, for billing and service maintenance (AKI.IO's privacy policy).

How much does Gemma 4 26B cost?

€0.168 per million input tokens and €0.84 per million output tokens, VAT included. A quick question — 300 tokens in, 400 out — costs about €0.0004. There is no subscription: credit starts at €2.

Can Gemma 4 26B call tools?

Yes: a real call through AKI.IO returned proper tool calls (function calling).

Which licence is Gemma 4 26B under?

Apache-2.0. Apache-2.0 is a permissive open-source licence: it allows commercial use.

Which checks has Gemma 4 26B passed?

6 of 6: Licence, Provenance, Security, GDPR, Benchmarks and Live test. Its passport shows the same stamps; how each check works.

Does Google see what I type?

No. Google published the weights, and AKI.IO runs them in Germany: your prompts go to AKI.IO, not to Google. AKI.IO adds that “model developers do not receive access to customer inference content through AKI.IO” (AKI.IO, read on 13 Oct 2026).