Gemma 4 26B A4B for JanitorAI: Free Proxy Setup on Composite
At a glance
| Model ID | google/gemma-4-26b-a4b-it |
|---|---|
| Free ID | google/gemma-4-26b-a4b-it:free |
| Made by | |
| Context window | 262,144 tokens |
| Longest reply | 235,929 tokens |
| Input | Text, images, video |
| Reasoning | Optional |
| Tool calling | Yes |
Use it as a JanitorAI proxy
- Create an API key on the API Keys page and link Discord on your account page (not needed with credits, a plan or your own provider key).
- In JanitorAI, open Settings → API, choose
OpenAIas the API type, and paste the full URLhttps://composite.lucidity.sh/v1/chat/completionsinto the OpenAI API URL field. - Paste the key, then type
google/gemma-4-26b-a4b-it:freeas the model name. - Send a short message to check the connection before you load a long chat.
Gemma 4 26B A4B IT is an instruction-tuned mixture-of-experts model from Google DeepMind. It has 25.2B parameters in total, but only 3.8B are active for each token, which Google says gives close to the quality of a 31B dense model at much lower cost. It reads text, images and video and has a 262,144-token window.
More calls per day
Because it is cheap to run, each call on the free ID takes a small share of the shared 80 a day, far less than the million-token Flash models. That makes it a reasonable everyday model for longer roleplay sessions on the free list. The Models page shows the current count per call.
Settings it accepts
It takes the full set of chat samplers: temperature, top P, top K, min P, repetition penalty, frequency and presence penalties, stop sequences and logit bias. A SillyTavern or JanitorAI preset applies as written. Reasoning is optional, with no effort levels listed, so control it in your app's settings rather than with a bracket suffix.
Compared with Gemma 3 4B
Gemma 3 4B is also free, but it is a small dense model better kept for short jobs like translation. Gemma 4 26B has twice the window, supports tool calling, and holds up better in long, layered scenes.
Other chat apps and the API
SillyTavern, ChubAI, Agnai and RisuAI take the same key and model name with the shorter base URL https://composite.lucidity.sh/v1.
Step-by-step setup for each app is in the JanitorAI and SillyTavern guides, and the docs hub covers the others.
From code, it is a normal OpenAI-compatible request:
curl https://composite.lucidity.sh/v1/chat/completions \
-H "Authorization: Bearer YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "google/gemma-4-26b-a4b-it:free", "messages": [{"role": "user", "content": "Hello"}]}'Free access (as of November 1, 2026)
On November 1, 2026, this model is on Composite's free list as google/gemma-4-26b-a4b-it:free. The plain google/gemma-4-26b-a4b-it stays paid. Every account shares 80 requests a day across all of the free models, separate from Synth's count, and the count resets at midnight UTC. Pricier free models count as several requests per call; the Models page shows how many. Credits and plans do not raise it.
The free list changes. Check the free models on the Models page for the current set, and the features guide for the rules. A :free id that is no longer on the list returns a model-not-found error and charges nothing. When the 80 requests are used up, requests stop with an error until the reset; Composite never charges your credits to continue. You can switch to google/gemma-4-26b-a4b-it on credits or a plan, or to Synth.