[BUG] Terminal output is not streaming (renders in chunks instead of token-by-token). But cursor cli can output by streaming.

Resolved 💬 7 comments Opened Feb 27, 2026 by zeevenn Closed Mar 31, 2026

Preflight Checklist

  • [x] I have searched existing issues and this hasn't been reported yet
  • [x] This is a single bug report (please file separate reports for different bugs)
  • [x] I am using the latest version of Claude Code

What's Wrong?

All text output in Claude Code CLI is not streaming token-by-token. Instead, text renders in buffered chunks — the terminal stays blank briefly, then a block of text appears at once.

This is especially noticeable (and frustrating) for the LAST text output in a turn: when tool calls are involved, the final response appears to wait until ALL content is fully generated before rendering anything, resulting in a long blank wait followed by a wall of text appearing instantly.

For shorter intermediate outputs the chunked rendering is less noticeable, but it's still not true token-by-token streaming like the Claude web interface provides.

What Should Happen?

All text output should stream token-by-token in real-time, similar to the Claude web interface, providing immediate visual feedback as the response is being generated.

The last text output should not wait for full completion before rendering.

Error Messages/Logs

2026-02-24T07:29:45.276Z [DEBUG] Stream started - received first chunk
2026-02-24T07:29:45.293Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:29:46.152Z [DEBUG] [useDeferredValue] Messages deferred by 1 (8→9)
2026-02-24T07:29:46.188Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:29:46.202Z [DEBUG] High write ratio: blit=0, write=1263 (100.0% writes), screen=31x148
2026-02-24T07:29:46.225Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:29:46.259Z [DEBUG] High write ratio: blit=0, write=1281 (100.0% writes), screen=32x148
2026-02-24T07:29:46.268Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:29:46.733Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:29:46.739Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:30:29.385Z [DEBUG] Fast mode unavailable: Fast mode is not available on Bedrock, Vertex, or Foundry
2026-02-24T07:30:29.944Z [WARN] Streaming stall detected: 43.2s gap between events (stall #1)

Steps to Reproduce

  1. Open a terminal (Ghostty, tested both with and without tmux)
  2. Run claude to start an interactive session
  3. Ask a question that requires tool calls followed by a text summary: "Read my ~/.zshrc and summarize what plugins I use"
  4. Observe:
  • Intermediate outputs render in chunks (not token-by-token)
  • The final summary text shows nothing for several seconds,

then the ENTIRE response appears at once

  1. Compare: ask a simple question like "What is 2+2?" — output still renders in chunks rather than streaming

Claude Model

application inference profile

Is this a regression?

I don't know

Last Working Version

_No response_

Claude Code Version

2.1.62 (Claude Code)

Platform

AWS Bedrock

Operating System

macOS

Terminal/Shell

Other

Additional Information

Terminal: ghostty

  • All text output is chunked, not just final responses
  • The final text output after tool calls is the worst case — appears to buffer the entire response before rendering
  • Tested in Ghostty directly (no tmux) — same behavior
  • Tested in tmux 3.6a — same behavior

View original on GitHub ↗

This issue has 7 comments on GitHub. Read the full discussion on GitHub ↗