[BUG] Opus 4.7 (1M context, Max plan) duplicates response output in terminal and consumes significantly more tokens than Opus 4.6
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?
When using Claude Opus 4.7 with the 1M context window (Max plan) in Claude Code, the response in the terminal appears to be duplicated or repeated. The same text (or something very close to it) is printed back multiple times within a single reply, making the output confusing and hard to read. It is not clear whether Claude is actually performing the same work more than once, spawning multiple sub-agents that each emit output, or simply re-printing the same final text — but the effect is that a single prompt produces output that looks like it has been multiplied.
In addition, token usage is substantially higher than with Opus 4.6 for comparable prompts. I suspect the duplication behavior may be contributing to the increased token consumption, but I cannot confirm whether the two issues are related.
The overall experience makes Opus 4.7 1M (Max) almost unusable for day-to-day work in Claude Code.
What Should Happen?
A single prompt should produce one clean, non-duplicated response in the terminal, and token usage should be broadly in line with what Opus 4.6 uses for a comparable task.
Error Messages/Logs
No explicit error messages are shown. The problem is visible in normal stdout — the assistant's response content itself is repeated. (Happy to attach a session transcript or --debug log if useful — let me know what would help.)
Steps to Reproduce
Launch Claude Code in the terminal.
Select the Claude Opus 4.7 model with the 1M context window on the Max plan.
Send any non-trivial prompt (the duplication appears to happen across a wide range of prompts, not a specific one).
Observe the response: the same content is repeated / multiplied within the single reply.
Compare token usage against the same prompt run on Opus 4.6 — 4.7 uses notably more tokens.
Claude Model
Opus
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.1.119
Platform
Anthropic API
Operating System
Windows
Terminal/Shell
VS Code integrated terminal
Additional Information
Does Opus 4.7 in Claude Code spawn multiple sub-agents or parallel workers whose outputs are being concatenated into a single visible reply?
If so, is that behavior expected only at the 1M context / Max tier, or is it present on all 4.7 configurations?
Could the elevated token usage vs. 4.6 be explained by the duplicated output, or is it an independent increase (e.g. more internal reasoning, more tool calls)?
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗