diff --git a/frontend/src/core/messages/utils.ts b/frontend/src/core/messages/utils.ts
index f2ea7f3eb..1024a0f50 100644
--- a/frontend/src/core/messages/utils.ts
+++ b/frontend/src/core/messages/utils.ts
@@ -106,6 +106,16 @@ export function getMessageGroups(messages: Message[]): MessageGroup[] {
     }
 
     if (message.type === "ai") {
+      // A message with answer content and no tool calls becomes its own
+      // assistant bubble below, which already renders the message's
+      // reasoning_content inside the bubble's <Reasoning> collapsible. Such a
+      // message must NOT also feed the processing group, or the ChainOfThought
+      // panel above the bubble paints the identical reasoning a second time
+      // (#3868). Intermediate reasoning (no content) and tool-calling steps
+      // still belong in the processing group.
+      const becomesAssistantBubble =
+        hasContent(message) && !hasToolCalls(message);
+
       if (hasPresentFiles(message)) {
         groups.push({
           id: message.id,
@@ -118,7 +128,10 @@ export function getMessageGroups(messages: Message[]): MessageGroup[] {
           type: "assistant:subagent",
           messages: [message],
         });
-      } else if (hasReasoning(message) || hasToolCalls(message)) {
+      } else if (
+        !becomesAssistantBubble &&
+        (hasReasoning(message) || hasToolCalls(message))
+      ) {
         const lastGroup = groups[groups.length - 1];
         // Accumulate consecutive intermediate AI messages into one processing group.
         if (lastGroup?.type !== "assistant:processing") {
@@ -132,9 +145,7 @@ export function getMessageGroups(messages: Message[]): MessageGroup[] {
         }
       }
 
-      // Not an else-if: a message with reasoning + content (but no tool calls) goes
-      // into the processing group above AND gets its own assistant bubble here.
-      if (hasContent(message) && !hasToolCalls(message)) {
+      if (becomesAssistantBubble) {
         groups.push({ id: message.id, type: "assistant", messages: [message] });
       }
     }
diff --git a/frontend/tests/unit/core/messages/usage.test.ts b/frontend/tests/unit/core/messages/usage.test.ts
index f400a7b98..90be6c6a8 100644
--- a/frontend/tests/unit/core/messages/usage.test.ts
+++ b/frontend/tests/unit/core/messages/usage.test.ts
@@ -42,7 +42,13 @@ test("counts later usage-bearing snapshots for the same AI message id", () => {
   });
 });
 
-test("keeps header and per-turn aggregation consistent for duplicated UI groups", () => {
+test("keeps header and per-turn aggregation consistent for a reasoning+answer message", () => {
+  // A single AI message carrying both reasoning (here via inline <think>) and
+  // answer text now lands in exactly one assistant group (#3868), so its usage
+  // is counted once both in the per-turn aggregation and against the header
+  // total. The by-id dedupe (see "accumulates each AI message usage only once")
+  // remains the defence-in-depth guard if any future grouping reintroduces a
+  // duplicate.
   const messages = [
     {
       id: "human-1",
@@ -61,15 +67,8 @@ test("keeps header and per-turn aggregation consistent for duplicated UI groups"
   const usageMessagesByGroupIndex = getAssistantTurnUsageMessages(groups);
   const turnUsageMessages = usageMessagesByGroupIndex.at(-1);
 
-  expect(groups.map((group) => group.type)).toEqual([
-    "human",
-    "assistant:processing",
-    "assistant",
-  ]);
-  expect(turnUsageMessages?.map((message) => message.id)).toEqual([
-    "ai-1",
-    "ai-1",
-  ]);
+  expect(groups.map((group) => group.type)).toEqual(["human", "assistant"]);
+  expect(turnUsageMessages?.map((message) => message.id)).toEqual(["ai-1"]);
   expect(accumulateUsage(messages)).toEqual(
     accumulateUsage(turnUsageMessages!),
   );
diff --git a/frontend/tests/unit/core/messages/utils.test.ts b/frontend/tests/unit/core/messages/utils.test.ts
index dde79e7ca..3c9ef6e0b 100644
--- a/frontend/tests/unit/core/messages/utils.test.ts
+++ b/frontend/tests/unit/core/messages/utils.test.ts
@@ -81,6 +81,76 @@ test("aggregates token usage messages once per assistant turn", () => {
   ).toEqual([null, null, ["ai-1", "ai-2"], null, ["ai-3"]]);
 });
 
+test("reasoning + content (no tool calls) yields a single assistant bubble, not a duplicate processing group", () => {
+  // Regression for #3868: in thinking/pro/ultra modes the final assistant
+  // message carries both reasoning_content and answer text. It must surface its
+  // reasoning exactly once — inside the assistant bubble's <Reasoning>
+  // collapsible. Routing the same message into a processing group as well makes
+  // the ChainOfThought panel above the bubble paint the identical reasoning a
+  // second time.
+  const messages = [
+    { id: "human-1", type: "human", content: "Why is the sky blue?" },
+    {
+      id: "ai-1",
+      type: "ai",
+      content: "Rayleigh scattering makes the sky blue.",
+      additional_kwargs: { reasoning_content: "Recall Rayleigh scattering." },
+    },
+  ] as Message[];
+
+  const groups = getMessageGroups(messages);
+
+  expect(groups.map((group) => group.type)).toEqual(["human", "assistant"]);
+
+  // The reasoning-bearing message lands in exactly one group, so turn-usage
+  // aggregation never double-counts it (see #2770).
+  const turnUsage = getAssistantTurnUsageMessages(groups);
+  expect(turnUsage.at(-1)?.map((message) => message.id)).toEqual(["ai-1"]);
+});
+
+test("keeps tool-call reasoning in the processing group while the final answer's reasoning rides its own bubble", () => {
+  // Companion to #3868: only the message that also becomes an assistant bubble
+  // (content, no tool calls) is pulled out of the processing group. Reasoning
+  // attached to an intermediate tool-calling step still belongs above, with its
+  // tool steps.
+  const messages = [
+    { id: "human-1", type: "human", content: "Search and summarize" },
+    {
+      id: "ai-1",
+      type: "ai",
+      content: "",
+      additional_kwargs: { reasoning_content: "I should search first." },
+      tool_calls: [{ id: "tool-1", name: "web_search", args: { query: "x" } }],
+    },
+    {
+      id: "tool-1-result",
+      type: "tool",
+      name: "web_search",
+      tool_call_id: "tool-1",
+      content: "[]",
+    },
+    {
+      id: "ai-2",
+      type: "ai",
+      content: "Here is the summary.",
+      additional_kwargs: { reasoning_content: "Synthesize the findings." },
+    },
+  ] as Message[];
+
+  const groups = getMessageGroups(messages);
+
+  expect(groups.map((group) => group.type)).toEqual([
+    "human",
+    "assistant:processing",
+    "assistant",
+  ]);
+  expect(groups[1]?.messages.map((message) => message.id)).toEqual([
+    "ai-1",
+    "tool-1-result",
+  ]);
+  expect(groups[2]?.messages.map((message) => message.id)).toEqual(["ai-2"]);
+});
+
 describe("inline <think> tag splitting", () => {
   test("strips a fully closed <think> block from AI content", () => {
     const message = aiMessage("<think>internal reasoning</think>final answer");