Models and Reasoning
The reasoning level sets how much the model thinks before it answers. A higher level produces more thorough answers, in exchange for longer response times and more credit usage.
Is a two-week sprint or Kanban a better fit for our 12-person product team?
Thought for 3s
For a 12-person product team, two-week sprints bring rhythm and predictability. But if there is a lot of unplanned work such as support requests, sprint plans break often, and Kanban becomes more efficient. Before deciding, look at the share of unplanned work over the last few weeks.
Write your message
| Level | Behavior | When to use |
|---|---|---|
| Low | Thinks briefly and answers quickly. Lowest latency and cost. | Simple questions and short texts |
| Medium | Balanced thinking. The recommended level for most work. | Most everyday work |
| High | Maximum thinking. | Hard problems, coding, and multi-step analysis |
- In the example above, the same question is answered at all three levels; as the level rises, thinking takes longer and the answer goes deeper.
- You can change the level at any time with the reasoning picker in the composer (see Interface).
- The default working mode sets the level new chats start with (see Default working mode).
Example
| Request | Recommended level |
|---|---|
| “Make this email more polite.” | Low |
| “Summarize this report and suggest three actions.” | Medium |
| “Find and fix the bug in this function.” | High |
Frequently asked questions
- Does reasoning affect credit usage? Yes. A higher level gives a more thorough response; in return, response time and credit usage go up.
- What level do scheduled tasks use? Medium reasoning (see Run profile).
- Can I change the level in the middle of a chat? Yes, from the reasoning selector in the composer, at any time.