diff --git a/cmd/tui/chat/chat.go b/cmd/tui/chat/chat.go
index 948105464..ef92f4ae0 100644
--- a/cmd/tui/chat/chat.go
+++ b/cmd/tui/chat/chat.go
@@ -224,9 +224,11 @@ func Run(ctx context.Context, opts Options) (*Result, error) {
 	m.nextImageID, m.nextAudioID = nextInputAttachmentIDsFromMessages(m.messages)
 	m.nextPastedTextID = nextInputPastedTextIDFromMessages(m.messages)
 	m.entries = entriesFromMessages(m.messages)
-	if !m.openModelOnInit {
-		m.refreshContextWindowTokens(m.opts.Model)
-	}
+	// Context window is resolved post-load (chatModelPreloadDoneMsg) rather than
+	// here: for local models /api/ps only reports the running num_ctx after the
+	// model loads, and opts.ContextWindowTokens already holds Show's max as a
+	// pre-load fallback. Refreshing now would just re-derive that same value
+	// (and block construction on a network call).
 	m.contextTokens = m.estimatePromptTokens(m.messages, "")
 	m.contextEstimate = true
 	if m.openModelOnInit {
@@ -375,7 +377,8 @@ func (m chatModel) Update(msg tea.Msg) (tea.Model, tea.Cmd) {
 			if msg.result.WorkingDir != "" {
 				m.workingDir = msg.result.WorkingDir
 			}
-			m.refreshContextWindowTokens(m.responseModelName(&msg.result.Latest))
+			// Context window is settled by preload (local num_ctx) or is
+			// static (cloud); no refresh needed post-run.
 			m.contextTokens = m.estimatePromptTokens(m.messages, "")
 			m.contextEstimate = true
 			if !messagesEndWithCompactionResult(m.messages) {
@@ -1093,7 +1096,6 @@ func (m *chatModel) startSkillRun(name, prompt string) (tea.Model, tea.Cmd) {
 }
 
 func (m *chatModel) startRunWithMessages(displayInput, historyInput string, newMessages []api.Message, extraSystemPrompt, skillName string) (tea.Model, tea.Cmd) {
-	m.refreshContextWindowTokens(m.opts.Model)
 	m.addPromptHistory(historyInput)
 	m.entries = append(m.entries, newChatEntry(chatEntry{role: "user", content: displayInput}))
 	if len(newMessages) > 1 {
diff --git a/cmd/tui/chat/events.go b/cmd/tui/chat/events.go
index 4e2f1be23..aa4df410b 100644
--- a/cmd/tui/chat/events.go
+++ b/cmd/tui/chat/events.go
@@ -109,7 +109,6 @@ func (m *chatModel) applyAgentEvent(event coreagent.Event) {
 		contextChanged = true
 	case coreagent.EventToolStarted:
 		m.resetStreamingState()
-		m.refreshContextWindowTokens(m.opts.Model)
 		startedAt := time.Now()
 		idx := m.findActiveToolEntry(event.ToolCallID)
 		if idx < 0 {
@@ -127,7 +126,6 @@ func (m *chatModel) applyAgentEvent(event coreagent.Event) {
 		m.markEntryDirty(idx)
 	case coreagent.EventToolFinished:
 		m.resetStreamingState()
-		m.refreshContextWindowTokens(m.opts.Model)
 		if event.WorkingDir != "" {
 			m.workingDir = event.WorkingDir
 		}
diff --git a/cmd/tui/chat/render.go b/cmd/tui/chat/render.go
index 862562adb..6cb21bced 100644
--- a/cmd/tui/chat/render.go
+++ b/cmd/tui/chat/render.go
@@ -1435,18 +1435,6 @@ func (m *chatModel) updateContextWindowTokens(tokens int) {
 	}
 }
 
-func (m chatModel) responseModelName(response *api.ChatResponse) string {
-	if response != nil {
-		if strings.TrimSpace(response.Model) != "" {
-			return response.Model
-		}
-		if strings.TrimSpace(response.RemoteModel) != "" {
-			return response.RemoteModel
-		}
-	}
-	return m.opts.Model
-}
-
 func (m chatModel) currentWorkingDir() string {
 	if strings.TrimSpace(m.workingDir) != "" {
 		return m.workingDir