Claude API 400: thinking.type.enabled is not supported for this model
Claude returns 400 saying thinking.type.enabled is not supported? Newer models removed extended thinking. Use adaptive thinking and output_config.effort.
Reviewed October 25, 2026
What it means in plain English
Anthropic's Claude API errors page says Claude 4.7 and later models have removed extended thinking. Sending thinking: {"type": "enabled"} to any of these models returns a 400 invalid_request_error:
"thinking.type.enabled" is not supported for this model. Use "thinking.type.adaptive" and "output_config.effort" to control thinking behavior.
The reverse also happens. Models that support only extended thinking, Claude 4.5 and earlier, reject thinking: {"type": "adaptive"} with:
adaptive thinking is not supported on this model
So the thinking setting has to match the model you call.
Likely causes
- You changed the model name to a newer Claude model, but the request still sends
thinking: {"type": "enabled", "budget_tokens": N}. - You moved to an older model, but the request sends
thinking: {"type": "adaptive"}. - A library or template sets the
thinkingfield for you.
How to fix it, step by step
- Find the request that sets the
thinkingfield, and note which model it calls. - On Claude 4.7 and later models, replace
thinking: {"type": "enabled"}with adaptive thinking. Anthropic says to use adaptive thinking, withoutput_config.effortto control how much thinking happens. Anthropic also says that if you omit thethinkingparameter, the request runs with adaptive thinking. - On Claude 4.5 and earlier models, use
thinking: {"type": "enabled", "budget_tokens": N}instead of adaptive. - Some newer models also reject
thinking: {"type": "disabled"}. If you get a 400 that namesthinking.type.disabled, read the message: Anthropic's errors page lists the setting each model accepts instead. - Redeploy and send a test request.
How to check it worked
- The request returns a normal response instead of a 400
invalid_request_error. - The
thinkingsetting matches the model, so switching models does not bring the error back.
Sources
Related fixes
More AI API errors fixes
Keep going on your own.
The guides stay free. A deep leak report is $15. Fix packs are $24, or $49 for all three. There is no done-for-you repair.