Skip to main content
Some replies need the model to think before it answers. Foundry supports reasoning models (GPT-5 family, newer Grok, others) and gives you two controls:
  1. Reasoning Effort — how deep should the model think?
  2. Live Reasoning View — watch the model think in real time.
Screenshot coming. Reasoning block above a reply.

​
When reasoning kicks in

Foundry auto-routes hard problems to reasoning-capable models. You’ll see a collapsible Reasoning block above the assistant’s reply when reasoning is active. The block stays collapsed by default. Click to expand and see the model’s chain of thought as it streams.

​
What the block contains

  • A Brain header with a spinner while thinking.
  • A streaming, monospace dump of the model’s reasoning trace.
  • A finished state once the answer is ready below.

​
Cost and time

Reasoning adds tokens and seconds. The deeper the effort, the bigger the bill and the longer the wait. Use Effort Control to dial it in.

Effort control

Low / Medium / High and what they buy you.

Live reasoning view

See the model think.