Bright runs on two model tiers. The composer’s model selector chooses between them.
  • Standard is the default tier, used whenever Genius is off.
  • Genius runs a more capable frontier model. The composer flags it while it is active, and it is the higher-cost setting.
The specific models behind each tier change over time. The tier behavior (default versus frontier, cheaper versus costlier) is the part worth designing your workflow around.

​
When to turn Genius on

  • Consequential or hard-to-reverse decisions, where a wrong answer is expensive to unwind.
  • Work with many interacting constraints that a weaker model tends to flatten.
  • Tasks where you cannot easily check the answer yourself.

​
When to leave it off

  • Lookups, navigation, status questions, and other bounded retrieval.
  • Mechanical edits and formatting.
  • Anything you will verify immediately anyway.
The Advisor specialist is a natural pairing for Genius-tier judgment: it exists to weigh plans and tradeoffs, the same situations where the stronger tier earns its cost.