anthropic: allow non-thinking models when using Anthropic API (#13692)

middleware/anthropic.go

@@ -118,6 +118,9 @@ func AnthropicMessagesMiddleware() gin.HandlerFunc {
 			return
 		}
 
+		// Set think to nil when being used with Anthropic API to connect to tools like claude code
+		c.Set("relax_thinking", true)
+
 		var b bytes.Buffer
 		if err := json.NewEncoder(&b).Encode(chatReq); err != nil {
 			c.AbortWithStatusJSON(http.StatusInternalServerError, anthropic.NewError(http.StatusInternalServerError, err.Error()))

middleware/anthropic_test.go

@@ -582,3 +582,26 @@ func TestAnthropicWriter_ErrorFromRoutes(t *testing.T) {
 		})
 	}
 }
+
+func TestAnthropicMessagesMiddleware_SetsRelaxThinkingFlag(t *testing.T) {
+	gin.SetMode(gin.TestMode)
+
+	var flagSet bool
+	router := gin.New()
+	router.Use(AnthropicMessagesMiddleware())
+	router.POST("/v1/messages", func(c *gin.Context) {
+		_, flagSet = c.Get("relax_thinking")
+		c.Status(http.StatusOK)
+	})
+
+	body := `{"model": "test-model", "max_tokens": 100, "messages": [{"role": "user", "content": "Hi"}]}`
+	req, _ := http.NewRequest(http.MethodPost, "/v1/messages", strings.NewReader(body))
+	req.Header.Set("Content-Type", "application/json")
+
+	resp := httptest.NewRecorder()
+	router.ServeHTTP(resp, req)
+
+	if !flagSet {
+		t.Error("expected relax_thinking flag to be set in context")
+	}
+}

server/routes.go

@@ -2072,8 +2072,14 @@ func (s *Server) ChatHandler(c *gin.Context) {
 		}
 	} else {
 		if req.Think != nil && req.Think.Bool() {
-			c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("%q does not support thinking", req.Model)})
-			return
+			// Set think to nil when being used with Anthropic API to connect to tools like claude code
+			if _, ok := c.Get("relax_thinking"); ok {
+				slog.Warn("model does not support thinking, relaxing thinking to nil", "model", req.Model)
+				req.Think = nil
+			} else {
+				c.JSON(http.StatusBadRequest, gin.H{"error": fmt.Sprintf("%q does not support thinking", req.Model)})
+				return
+			}
 		}
 	}