Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 18 additions & 6 deletions build/agents/build-your-agent/agent-settings/language-model.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -52,14 +52,26 @@ Adjust based on whether you need brief updates or comprehensive explanations fro

### Reasoning / Thinking

**What it is:** For supported models (OpenAI, Google, Anthropic), you can configure your models to use output tokens to 'reason'. Decisions will be slower and cost more, but it improves performance on complex problems. Ignored unless your selected agent model supports reasoning and you choose a consistent config, e.g. use OpenAI reasoning effort for OpenAI o-series models.
For supported models, you can configure how much the model reasons before responding. Reasoning uses output tokens to work through complex problems, which improves quality on hard tasks but is slower and costs more.

**How to set it:**
#### Thinking type and effort level

For supported Claude models, the Thinking setting has two controls: set **Thinking Type** to **Adaptive** to enable reasoning, and use **Effort Level** (Low, Medium, or High — up to Max on some models) to balance quality against speed and cost.

<img
src="/images/agent/advanced-settings-thinking.png"
/>

To reach these settings, go to **Agent → Settings → Advanced settings → Language Model**.

Using the 'Provider' dropdown, set this to match the model you've chosen for your Agent (OpenAI, Google, Anthropic)
- For **OpenAI**, you can then select 'Reasoning Effort'. The default is Medium, and you can also select Low or High.
- For **Google**, you can enter a 'Thinking Budget', which can be set at any positive value to enable thinking
- For **Anthropic**, you can enter a 'Thinking Budget', which can be set at any value above 1024 to enable thinking
These controls appear on:

- Claude Opus 4.6, 4.7, and 4.8
- Claude Sonnet 4.6

<Warning>
Thinking tokens count as output tokens, not a separate category, so enabling reasoning increases your cost. Monitor usage when you turn it on.
</Warning>

---

Expand Down
Binary file added images/agent/advanced-settings-thinking.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Loading