the models
The Axion model family
Built, safety-tested, and hosted by us. Use them in the browser or the API.
Lumen 1.2.5
Our flagship. A fine-tuned Llama 3.1 8B with two-stage training — supervised fine-tuning followed by DPO safety alignment. Best for chat, questions, writing, and code explanations.
Veil 1.1
Our first and fastest model — a fine-tuned Llama 3.1 3B for quick general chat.
the api
One curl away
The API speaks the OpenAI chat-completions format, so every SDK and tool that works with OpenAI works with Lumen — just change the base URL.
$ curl https://api.amplifiedsmp.org/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $AXION_API_KEY" \
-d '{"model": "lumen", "messages": [{"role": "user", "content": "hello!"}]}'
ways to use it
Wherever you work
The same models, three doors in.
Chat
A clean chat interface for Lumen. Signed-in chats sync across your devices.
POST /v1/chat/completionsAPI
An OpenAI-compatible endpoint, so every SDK and tool that already speaks OpenAI works by changing the base URL.
$ npm i -g @axion-labs-ai/quark-cliAxion CLI
Our open-source terminal agent — it edits files, runs commands, and uses Lumen (or any model you bring) as its brain.
safety
We publish our safety work
Every Lumen release ships with adversarial testing — and when a model fails, we take it down and say exactly why.
Lumen 1.2.5
Safety report: 73% pass rate
DPO safety training results across 15 adversarial scenarios.
Lumen 1.2.1 · June 2026
Why we suspended a model
Testing showed dangerous self-preservation behavior. We took it offline and published the findings.
Lumen 1.2.1 · June 2026
Safety report: 26% pass rate
The full 1.2.1 results — 7 scenarios, 5 runs each, every prompt and response published.
Start talking to Lumen
Create a free Axion account, then use Lumen in the browser, API, or terminal.
Create account & chat Get API key