Models: Atlarix Core and your own keys

Atlarix Core is the managed model catalogue in the app: sign in, pick one of the included models, and start building — no vendor API key required for Core itself. Models are shown by their real names in the picker.

Every feature is on every plan: unlimited workspaces, both marketplaces, one-click connectors, automations. Your own API keys and local models cost nothing, ever. What you pay for is Atlarix Core, the managed models we host and meter — top up from $10 and it is billed as you use it. Pro includes Core for everyday work and adds Atlarix Reviewer, drawing on the same single balance.

Each Core model carries its own context window, and the app shows the live figure in the chat meter:

• Core 1 — GLM-5.3 (z-ai/glm-5.3) — 1.0M context, supports deep thinking
• Core 2 — GPT-5.6 Terra (openai/gpt-5.6-terra) — 1.1M context, supports deep thinking
• Core 3 — GPT-6 Astra (openai/gpt-6-astra) — 1.1M context, supports deep thinking

The lineup is configured remotely and can change without an app update, so treat this as what is mapped today rather than a permanent list.

There is no tier gating. Every plan can use every Core model, chosen per request from the picker. What differs is who has already paid for the inference: Pro includes Atlarix Core for everyday work, and heavier use draws on the same pay-as-you-go credit Free uses from the start.

Add provider keys under Settings → AI: 210+ providers and 6,500+ tool-capable models from the live models.dev catalogue, including OpenAI, Anthropic, Gemini, and any OpenAI-compatible endpoint.

Your keys and Atlarix Core coexist — pick the active model per chat and switch anytime without leaving the workspace. Turns on your own key show an estimated cost per reply.

Your own keys and local models are free on any plan.

Ollama and LM Studio work the same way as any other provider — select the model in the picker and it runs on your machine.

Models that can reason expose a Deep thinking toggle in the chat config dropdown, next to the model picker. It only appears when the selected model actually supports reasoning — there is no switch to find on a model that cannot use it.

Off is not "no thinking": the default is already high reasoning effort. On asks for the maximum that particular model supports, and it scales to each model rather than sending one fixed setting everywhere.

It applies to Atlarix Core and to your own keys alike.

Atlarix prompts you to top up or switch to another model, and keeps your conversation intact. Nothing is lost.