Problem
When Claude Code enters extended thinking mode, the terminal only displays thinking... with no further output. During long thinking phases (3-5+ minutes), it's impossible for users to tell whether:
- The model is genuinely reasoning through a complex problem, or
- The Anthropic API is hanging/experiencing an outage (which happens regularly)
This creates uncertainty and wastes time — users end up waiting for a response that may never come, or canceling a request that was actually making progress.
Current behavior
thinking... is shown as a static label with no content- The
redact-thinking API flag (introduced ~v2.1.68) strips thinking content server-side, so even with showThinkingSummaries: true and alwaysThinkingEnabled: true, there's nothing to display
Ctrl+O toggles verbose mode but doesn't help since the thinking content itself is redacted- Elapsed time is shown and counts up, but that alone doesn't confirm the model is still actively processing
Proposed solution
Stream thinking summaries in real-time during the thinking phase, similar to how response tokens are streamed. Even short periodic summaries (e.g., "Analyzing code structure...", "Evaluating approach...") would give users confidence that the model is still active and making progress.
Alternatively, at minimum show additional progress indicators (token count, heartbeat/activity signal) alongside the existing elapsed timer so users can distinguish an active thinking session from a stalled API connection.
Impact
This is a significant UX issue for users working with complex tasks where extended thinking is common. The uncertainty leads to unnecessary cancellations and erodes trust in the tool's reliability.
Problem
When Claude Code enters extended thinking mode, the terminal only displays
thinking...with no further output. During long thinking phases (3-5+ minutes), it's impossible for users to tell whether:This creates uncertainty and wastes time — users end up waiting for a response that may never come, or canceling a request that was actually making progress.
Current behavior
thinking...is shown as a static label with no contentredact-thinkingAPI flag (introduced ~v2.1.68) strips thinking content server-side, so even withshowThinkingSummaries: trueandalwaysThinkingEnabled: true, there's nothing to displayCtrl+Otoggles verbose mode but doesn't help since the thinking content itself is redactedProposed solution
Stream thinking summaries in real-time during the thinking phase, similar to how response tokens are streamed. Even short periodic summaries (e.g.,
"Analyzing code structure...","Evaluating approach...") would give users confidence that the model is still active and making progress.Alternatively, at minimum show additional progress indicators (token count, heartbeat/activity signal) alongside the existing elapsed timer so users can distinguish an active thinking session from a stalled API connection.
Impact
This is a significant UX issue for users working with complex tasks where extended thinking is common. The uncertainty leads to unnecessary cancellations and erodes trust in the tool's reliability.