diff --git a/internal/multiagent/eino_summarize.go b/internal/multiagent/eino_summarize.go
index b7cd7e413..38647f1ab 100644
--- a/internal/multiagent/eino_summarize.go
+++ b/internal/multiagent/eino_summarize.go
@@ -300,9 +300,13 @@ func newEinoSummarizationModelOptions(outputReserve int, modelName, kind string,
 	if strings.TrimSpace(kind) != "" && kind != "classic" {
 		label = "eino " + kind + " summarization generate request"
 	}
+	tokenLimit := einoopenai.WithMaxCompletionTokens(outputReserve)
+	if oa != nil && isEinoAgenticClaudeProvider(oa.Provider) {
+		// Native Claude consumes the common option; OpenAI rejects both token fields together.
+		tokenLimit = model.WithMaxTokens(outputReserve)
+	}
 	return []model.Option{
-		model.WithMaxTokens(outputReserve),
-		einoopenai.WithMaxCompletionTokens(outputReserve),
+		tokenLimit,
 		einoopenai.WithExtraHeader(map[string]string{
 			copenai.SummarizationRequestHeader: "1",
 		}),
diff --git a/internal/multiagent/eino_summarize_payload_test.go b/internal/multiagent/eino_summarize_payload_test.go
index 4b6890d10..0ea7ceaf6 100644
--- a/internal/multiagent/eino_summarize_payload_test.go
+++ b/internal/multiagent/eino_summarize_payload_test.go
@@ -93,9 +93,9 @@ func TestStripReasoningFromSummarizationPayloadHonorsOpenAICompatProfileForNonDe
 	}
 }
 
-func TestEinoSummarizationModelOptionsSetCommonMaxTokens(t *testing.T) {
+func TestEinoSummarizationModelOptionsSetClaudeCommonMaxTokens(t *testing.T) {
 	const outputReserve = 4096
-	opts := newEinoSummarizationModelOptions(outputReserve, "minimax-m3", "agentic", nil, nil)
+	opts := newEinoSummarizationModelOptions(outputReserve, "claude-sonnet", "agentic", &config.OpenAIConfig{Provider: "claude"}, nil)
 	common := model.GetCommonOptions(nil, opts...)
 	if common == nil || common.MaxTokens == nil {
 		t.Fatal("expected summarization options to set common max_tokens")
diff --git a/internal/multiagent/eino_summarize_tokens_test.go b/internal/multiagent/eino_summarize_tokens_test.go
new file mode 100644
index 000000000..da118ec51
--- /dev/null
+++ b/internal/multiagent/eino_summarize_tokens_test.go
@@ -0,0 +1,113 @@
+package multiagent
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"testing"
+
+	"cyberstrike-ai/internal/config"
+
+	"github.com/cloudwego/eino-ext/components/model/agenticopenai"
+	einoopenai "github.com/cloudwego/eino-ext/components/model/openai"
+	"github.com/cloudwego/eino/schema"
+)
+
+func TestEinoSummarizationSendsOneTokenLimit(t *testing.T) {
+	for _, kind := range []string{"classic", "agentic"} {
+		t.Run(kind, func(t *testing.T) {
+			const outputReserve = 4096
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				var body map[string]json.RawMessage
+				if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+					t.Error(err)
+					w.WriteHeader(http.StatusBadRequest)
+					return
+				}
+				w.Header().Set("Content-Type", "application/json")
+				if _, exists := body["max_tokens"]; exists {
+					t.Errorf("summary request includes max_tokens alongside max_completion_tokens: %s", body["max_tokens"])
+					w.WriteHeader(http.StatusBadRequest)
+					_, _ = w.Write([]byte(`{"error":{"message":"max_tokens and max_completion_tokens cannot be set at the same time","type":"invalid_request_error"}}`))
+					return
+				}
+				if string(body["max_completion_tokens"]) != "4096" {
+					t.Errorf("summary output budget = %s, want 4096", body["max_completion_tokens"])
+				}
+				_, _ = w.Write([]byte(`{"id":"summary","object":"chat.completion","model":"test-model","choices":[{"index":0,"message":{"role":"assistant","content":"summary"},"finish_reason":"stop"}]}`))
+			}))
+			defer server.Close()
+
+			ctx := context.Background()
+			oa := &config.OpenAIConfig{Provider: "openai_compatible", Model: "test-model"}
+			opts := newEinoSummarizationModelOptions(outputReserve, oa.Model, kind, oa, nil)
+			defaultLimit := 8192
+			if kind == "classic" {
+				chat, err := einoopenai.NewChatModel(ctx, &einoopenai.ChatModelConfig{
+					APIKey: "test", BaseURL: server.URL, Model: oa.Model, HTTPClient: server.Client(), MaxCompletionTokens: &defaultLimit,
+				})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := chat.Generate(ctx, []*schema.Message{schema.UserMessage("summarize")}, opts...); err != nil {
+					t.Fatal(err)
+				}
+			} else {
+				chat, err := agenticopenai.NewChatModel(ctx, &agenticopenai.ChatConfig{
+					APIKey: "test", BaseURL: server.URL, Model: oa.Model, HTTPClient: server.Client(), MaxCompletionTokens: &defaultLimit,
+				})
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := chat.Generate(ctx, []*schema.AgenticMessage{schema.UserAgenticMessage("summarize")}, opts...); err != nil {
+					t.Fatal(err)
+				}
+			}
+		})
+	}
+}
+
+func TestEinoSummarizationPreservesClaudeTokenLimit(t *testing.T) {
+	for _, kind := range []string{"classic", "agentic"} {
+		t.Run(kind, func(t *testing.T) {
+			server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+				var body map[string]json.RawMessage
+				if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
+					t.Error(err)
+					w.WriteHeader(http.StatusBadRequest)
+					return
+				}
+				if string(body["max_tokens"]) != "4096" {
+					t.Errorf("Claude summary budget = %s, want 4096", body["max_tokens"])
+				}
+				if _, exists := body["max_completion_tokens"]; exists {
+					t.Error("Claude request includes OpenAI token limit")
+				}
+				w.Header().Set("Content-Type", "application/json")
+				_, _ = w.Write([]byte(`{"id":"summary","type":"message","role":"assistant","model":"claude-sonnet","content":[{"type":"text","text":"summary"}],"stop_reason":"end_turn","usage":{"input_tokens":1,"output_tokens":1}}`))
+			}))
+			defer server.Close()
+			ctx := context.Background()
+			oa := config.OpenAIConfig{Provider: "claude", APIKey: "test", BaseURL: server.URL, Model: "claude-sonnet", MaxCompletionTokens: 8192}
+			opts := newEinoSummarizationModelOptions(4096, oa.Model, kind, &oa, nil)
+			if kind == "classic" {
+				chat, err := newEinoToolCallingChatModelFactory(server.Client(), nil, nil)(ctx, oa, einoModelModeNormal)
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := chat.Generate(ctx, []*schema.Message{schema.UserMessage("summarize")}, opts...); err != nil {
+					t.Fatal(err)
+				}
+			} else {
+				chat, err := newEinoAgenticChatModelFactory(server.Client(), nil, nil)(ctx, oa, einoModelModeNormal)
+				if err != nil {
+					t.Fatal(err)
+				}
+				if _, err := chat.Generate(ctx, []*schema.AgenticMessage{schema.UserAgenticMessage("summarize")}, opts...); err != nil {
+					t.Fatal(err)
+				}
+			}
+		})
+	}
+}