You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Thanks for the request. This is available in 2026.10.29.
Pin the reasoning effort with a routing rule. Open Routing, edit the rule that forwards Claude Code to GLM, and under Parameter rewrites set Reasoning to max. Every request that rule matches is then sent with max, whatever the client asks for, including the xhigh that Ultracode always sends. The level applies to requests that already have reasoning on; requests without reasoning (such as title generation) are left unchanged. Off turns reasoning off.
If the rule can fail over to another upstream and only GLM should get max, add a separate rule with the condition Selected upstream = your GLM upstream, On match = Continue matching, and Reasoning set there; such a rule applies only to that upstream.
Unsupported levels are also handled without any setting. When an upstream answers 400 because it does not accept the requested level, ThinkWatch sends the request again once with a level it accepts (the one its error message lists, or the model's default), and remembers that for the model for 7 days, so later requests go out correctly the first time. Every change is shown in the request details.
Activity
fylorn commentedon Oct 25, 2026
Thanks for the request. This is available in 2026.10.29.
Pin the reasoning effort with a routing rule. Open Routing, edit the rule that forwards Claude Code to GLM, and under Parameter rewrites set Reasoning to
max. Every request that rule matches is then sent withmax, whatever the client asks for, including thexhighthat Ultracode always sends. The level applies to requests that already have reasoning on; requests without reasoning (such as title generation) are left unchanged.Offturns reasoning off.If the rule can fail over to another upstream and only GLM should get
max, add a separate rule with the condition Selected upstream = your GLM upstream, On match = Continue matching, and Reasoning set there; such a rule applies only to that upstream.Unsupported levels are also handled without any setting. When an upstream answers 400 because it does not accept the requested level, ThinkWatch sends the request again once with a level it accepts (the one its error message lists, or the model's default), and remembers that for the model for 7 days, so later requests go out correctly the first time. Every change is shown in the request details.