All docs · The agent
Model and reasoning
The engine that powers the agent. Fast models are included; the most powerful ones activate with your own provider key.
On this page
At the top of the editor, the model chip opens the list of models, arranged in two sections. Here is exactly what you see:
Included: nothing to configure
The fast models (GPT-5.4 Nano and Mini, Claude Haiku, Gemini Flash) run on Magnitude’s keys. They are enough for the vast majority of conversations: they are ready out of the box, with no key and no extra cost.
Premium: your own key
The most powerful models (GPT-5.4, Claude Sonnet and Opus, Gemini Pro) are activated by connecting the API key from your own OpenAI, Anthropic or Google account, in Settings › AI Models. You are billed directly by the provider, at the listed rate (input / output per million tokens). Until a key is added, these models stay locked.
A key you add is never shown again: the system only indicates whether a provider is connected, never the key itself.
The reasoning level
On the models that support it (the GPT-5 family), a slider sets the reasoning level: from "fast" to "thorough". The higher it is, the more time the agent takes to reason through complex cases, at the cost of a little latency.