Build

Thinking levels

Choose how hard Kael works on a request with reasoning_effort.

Levels

The thinking level changes how Kael works on a request, not only how long it thinks. Start at low, the default, and move up when an answer is not good enough.

Level	Use it for	Price per 1M tokens
`off`	The fastest replies, when a question does not need extra thinking.	$2.00 in, $6.00 out
`low` (default)	Everyday questions, quick edits, short answers.	$2.00 in, $6.00 out
`high`	Debugging, code review, and problems with several steps.	$2.00 in, $6.00 out
`max`	Your hardest work: security reviews, large refactors, anything you will act on without checking.	$2.00 in, $6.00 out
`z-low`, `z-high`	The most demanding work, at a higher price.	$6.00 in, $15.00 out

Off is the fastest. The first word usually arrives within a few seconds. At every other level Kael thinks first, so the wait is longer and the output then streams quickly.
Off, low, high and max cost the same per token. Higher levels think for longer, so a request takes more time and can use more tokens. z-low and z-high are priced higher per token. See Pricing and billing.
The level applies to the whole request, both the draft and the review.

Set the level

The setting has a different name in each format, but the same six values.

curl https://api.quancis.space/v1/chat/completions \
  -H "Authorization: Bearer $QUAN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kael-beta",
    "reasoning_effort": "high",
    "messages": [{"role": "user", "content": "Find the race condition in this code: ..."}]
  }'

Values are not case sensitive. With the OpenAI Python SDK or the Anthropic SDK, pass the field through extra_body so it works whatever the SDK's own type list says.

What is not available

There is no auto level, so choose a level yourself. medium does not exist either. Any value outside the six returns a 400 that names the field, so a typo never runs at a level you did not choose.

400 response

{
  "error": {
    "message": "Invalid reasoning_effort 'auto'. Kael Beta supports 'off', 'low', 'high', 'max', 'z-low', or 'z-high'.",
    "type": "invalid_request_error",
    "param": "reasoning_effort"
  }
}

The Messages format's own thinking field is accepted and ignored.

In the Playground

The Playground has a thinking setting too, and remembers your choice in this browser.