happened with kimi k2 thinking for me a few times too
Mark Florkowski
Team
Reasoning models can sometimes do this if they use all of their token budget during reasoning. We are working on making this less likely to occur/better feedback when it does. Enabling search makes this more likely to occur.
Aquamarine Cookie
Thanks for the explanation. Do you have any recommendations on how to proceed at this time if our use case requires both reasoning and search enabled?
Mark Florkowski
Team
use a different model to search, then switch back to reasoning model afterwards for analysis. Or use lower reasoning levels
Log in to comment and vote
Comments4
Coral Cucumber
happened with kimi k2 thinking for me a few times too
Mark Florkowski
Reasoning models can sometimes do this if they use all of their token budget during reasoning. We are working on making this less likely to occur/better feedback when it does. Enabling search makes this more likely to occur.
Aquamarine Cookie
Thanks for the explanation. Do you have any recommendations on how to proceed at this time if our use case requires both reasoning and search enabled?
Mark Florkowski
use a different model to search, then switch back to reasoning model afterwards for analysis. Or use lower reasoning levels