Choosing a model
On Agent, open the model picker in the chat input bar to select the hosted model for the current brand. Connect does not use this picker: choose and pay for the model in your external assistant instead.
Current hosted models by relative cost tier
- $ — Gemini 2.5 Flash, Claude Haiku 4.5, GPT-5.6 Luna
- $$ — Gemini 3.1 Pro, GPT-5.4, GPT-5.5, GPT-5.6 Terra
- $$$ — GPT-5.6 Sol, Claude Sonnet 4.6, Claude Sonnet 5, Claude Opus 4.6, 4.7, and 4.8
- $$$$ — Claude Fable 5
The symbols are relative, not fixed fees. Actual credits depend on input, output, reasoning, and tool use. The live picker is authoritative if this list changes between documentation reviews.
Defaults — paid Agent brands inherit GPT-5.6 Luna at X-High effort unless an admin changes the platform default. The free trial is fixed to GPT-5.6 Luna at High effort and cannot switch models or effort.
Reasoning effort — the GPT-5.6 family and adaptive-thinking Claude models show a second control for Auto, Low, Medium, High, X-High, or Max. A change applies to the next turn in the current chat and carries forward to new chats for the same brand. Higher effort can improve difficult work but usually takes longer and consumes more credits. Provider support varies, so Magister maps unsupported levels safely.
Per-brand choice — model and effort preferences are saved per brand. Use faster $ models for simple, tool-heavy work; Terra, Gemini Pro, GPT-5.4, or GPT-5.5 for balanced work; and Sol, Sonnet, Opus, or Fable when quality matters more than cost.
BYOK — organization admins can add Anthropic or OpenAI keys for hosted Agent projects, including hosted Agent add-ons in Connect organizations. Eligible requests use that provider account instead of Magister model credits. Standalone Connect uses its external assistant's credentials, and Free cannot save or use provider keys. See Bring Your Own Key.