diff --git a/ChatView/ChatRootView.cpp b/ChatView/ChatRootView.cpp
index 6a22701..a57b5ac 100644
--- a/ChatView/ChatRootView.cpp
+++ b/ChatView/ChatRootView.cpp
@@ -131,6 +131,7 @@ ChatRootView::ChatRootView(QQuickItem *parent)
 
     connect(m_chatModel, &ChatModel::modelReseted, this, [this]() {
         setRecentFilePath(QString{});
+        m_tokenCounter->resetServerUsage();
         m_fileEditController->clearCurrentRequestId();
     });
     auto maybeEmitTitle = [this] {
@@ -231,8 +232,8 @@ ChatRootView::ChatRootView(QQuickItem *parent)
         m_clientInterface,
         &ClientInterface::messageUsageReceived,
         this,
-        [this](int promptTokens, int /*completionTokens*/, int /*cached*/, int /*reasoning*/) {
-            m_tokenCounter->recordServerUsage(promptTokens);
+        [this](int promptTokens, int /*completionTokens*/, int cachedTokens, int /*reasoning*/) {
+            m_tokenCounter->recordServerUsage(promptTokens, cachedTokens);
         });
 
     connect(
@@ -489,8 +490,6 @@ void ChatRootView::dispatchSend(
         }
     }
 
-    m_tokenCounter->recordSent();
-
     if (currentChatAgent().isEmpty())
         loadAvailableChatAgents();
 
diff --git a/ChatView/InputTokenCounter.cpp b/ChatView/InputTokenCounter.cpp
index 27c0794..05f9573 100644
--- a/ChatView/InputTokenCounter.cpp
+++ b/ChatView/InputTokenCounter.cpp
@@ -4,9 +4,6 @@
 
 #include "InputTokenCounter.hpp"
 
-#include <algorithm>
-
-#include "Logger.hpp"
 #include "context/ContextManager.hpp"
 #include "context/TokenUtils.hpp"
 
@@ -49,8 +46,6 @@ void InputTokenCounter::setLinkedFiles(const QStringList &linkedFiles)
 
 void InputTokenCounter::recompute()
 {
-    int inputTokens = m_messageTokens;
-
     const auto splitImageEstimate = [](const QStringList &paths, QStringList &textPaths) {
         int imageTokens = 0;
         for (const QString &p : paths) {
@@ -62,15 +57,24 @@ void InputTokenCounter::recompute()
         return imageTokens;
     };
 
+    int pendingTokens = m_messageTokens;
     if (!m_attachments.isEmpty()) {
         QStringList textPaths;
-        inputTokens += splitImageEstimate(m_attachments, textPaths);
+        pendingTokens += splitImageEstimate(m_attachments, textPaths);
         if (!textPaths.isEmpty()) {
             auto attachFiles = m_contextManager->getContentFiles(textPaths);
-            inputTokens += Context::TokenUtils::estimateFilesTokens(attachFiles);
+            pendingTokens += Context::TokenUtils::estimateFilesTokens(attachFiles);
         }
     }
 
+    if (m_hasServerUsage && m_history && !m_history->isEmpty()) {
+        m_inputTokens = m_serverInputTokens + pendingTokens;
+        emit inputTokensChanged();
+        return;
+    }
+
+    int inputTokens = pendingTokens;
+
     if (!m_linkedFiles.isEmpty()) {
         QStringList textPaths;
         inputTokens += splitImageEstimate(m_linkedFiles, textPaths);
@@ -87,33 +91,25 @@ void InputTokenCounter::recompute()
         }
     }
 
-    m_inputTokens = static_cast<int>(inputTokens * m_calibrationFactor);
+    m_inputTokens = inputTokens;
     emit inputTokensChanged();
 }
 
-void InputTokenCounter::recordSent()
+void InputTokenCounter::recordServerUsage(int promptTokens, int cachedTokens)
 {
-    m_lastSentEstimate = m_calibrationFactor > 0.0
-                             ? static_cast<int>(m_inputTokens / m_calibrationFactor)
-                             : m_inputTokens;
-}
-
-void InputTokenCounter::recordServerUsage(int promptTokens)
-{
-    if (promptTokens <= 0 || m_lastSentEstimate <= 0)
+    const int serverInput = promptTokens + cachedTokens;
+    if (serverInput <= 0)
         return;
 
-    const double rawFactor
-        = static_cast<double>(promptTokens) / static_cast<double>(m_lastSentEstimate);
-    const double clamped = std::clamp(rawFactor, 0.5, 3.0);
-    m_calibrationFactor = 0.5 * m_calibrationFactor + 0.5 * clamped;
-
-    LOG_MESSAGE(QString("Token calibration: server=%1 estimated=%2 ratio=%3 ema=%4")
-                    .arg(promptTokens)
-                    .arg(m_lastSentEstimate)
-                    .arg(rawFactor, 0, 'f', 3)
-                    .arg(m_calibrationFactor, 0, 'f', 3));
+    m_serverInputTokens = serverInput;
+    m_hasServerUsage = true;
+    recompute();
+}
 
+void InputTokenCounter::resetServerUsage()
+{
+    m_serverInputTokens = 0;
+    m_hasServerUsage = false;
     recompute();
 }
 
diff --git a/ChatView/InputTokenCounter.hpp b/ChatView/InputTokenCounter.hpp
index 9f79982..4e7d366 100644
--- a/ChatView/InputTokenCounter.hpp
+++ b/ChatView/InputTokenCounter.hpp
@@ -34,8 +34,8 @@ public:
     void setLinkedFiles(const QStringList &linkedFiles);
     void recompute();
 
-    void recordSent();
-    void recordServerUsage(int promptTokens);
+    void recordServerUsage(int promptTokens, int cachedTokens);
+    void resetServerUsage();
 
 signals:
     void inputTokensChanged();
@@ -48,8 +48,8 @@ private:
     QStringList m_linkedFiles;
     int m_messageTokens{0};
     int m_inputTokens{0};
-    int m_lastSentEstimate{0};
-    double m_calibrationFactor{1.0};
+    int m_serverInputTokens{0};
+    bool m_hasServerUsage{false};
 };
 
 } // namespace QodeAssist::Chat