rc upds

th3ist.go

@@ -117,6 +117,7 @@ type ChatCompletionRequest struct {
   MaxCompletionTokens int           `json:"max_completion_tokens"`
   Temperature         *float64      `json:"temperature,omitempty"`
   TopP                *float64      `json:"top_p,omitempty"`
+  ReasoningEffort     string        `json:"reasoning_effort,omitempty"`
 }
 
 type ChatCompletionResponseChoice struct {
@@ -2418,6 +2419,11 @@ func (g *T3Gateway) HandleChatCompletions(w http.ResponseWriter, r *http.Request
     })
   }
 
+  effReasoningEffort := "low"
+  if req.ReasoningEffort != "" {
+    effReasoningEffort = req.ReasoningEffort
+  }
+
   t3Payload := T3ChatPayload{
     Messages: t3Messages,
     ThreadMetadata: T3ThreadMetadata{
@@ -2431,7 +2437,7 @@ func (g *T3Gateway) HandleChatCompletions(w http.ResponseWriter, r *http.Request
     Model:             modelName,
     ConvexSessionID:   sessionID,
     ModelParams: T3ModelParams{
-      ReasoningEffort: "low",
+      ReasoningEffort: effReasoningEffort,
       IncludeSearch:   false,
       SearchLimit:     1,
     },
@@ -2440,7 +2446,7 @@ func (g *T3Gateway) HandleChatCompletions(w http.ResponseWriter, r *http.Request
       CreationTime: nowMs,
       CurrentModelParameters: T3UserConfigParams{
         IncludeSearch:   false,
-        ReasoningEffort: "low",
+        ReasoningEffort: effReasoningEffort,
       },
       CurrentlySelectedModel: modelName,
       LatestTOSDate:          int64(nowMs),