[BUG] 400 Input is too long for requested model with `global.anthropic.claude-sonnet-4-5-20250929-v1:0`
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?
I've never had this issue with any previous sonnet bedrock models. But ever since using global.anthropic.claude-sonnet-4-5-20250929-v1:0 on Claude-Code v2.0.1 I'm constantly seeing the error:
API Error: 400 Input is too long for requested model.
Even trying to do a force compaction doesn't solve the issue.
Running maocs 15.6.1
What Should Happen?
It should never throw a 400 error, and compact if it's running out of context length.
Error Messages/Logs
API Error: 400 Input is too long for requested model.
Steps to Reproduce
Run claude-code with the following environment variables set:
AWS_PROFILE=sso-caas-dev-private CLAUDE_CODE_USE_BEDROCK=1 ANTHROPIC_MODEL='global.anthropic.claude-sonnet-4-5-20250929-v1:0' ANTHROPIC_SMALL_FAST_MODEL='us.anthropic.claude-3-5-haiku-20241022-v1:0' AWS_REGION=us-west-2 claude
Keep running the prompts until you get close to the end of the context. You'll start to see this error message:
API Error: 400 Input is too long for requested model.
Trying to run /compact - even when the compaction occurs doesn't solve the issue.
Claude Model
None
Is this a regression?
Yes, this worked in a previous version
Last Working Version
_No response_
Claude Code Version
2.0.1
Platform
AWS Bedrock
Operating System
macOS
Terminal/Shell
iTerm2
Additional Information
<img width="366" height="516" alt="Image" src="https://github.com/user-attachments/assets/ab078430-20a6-4008-ab1c-e0e17877fcaf" />
18 Comments
I think the issue is that those log files I was loading in were very large, and weren't able to be compacted away, or filled the context. My work around was to tell clause that those files were large so it would need to be smart about reading them.
This happens on every MCP almost that goes the max limit, why can't the max limit be respected by CC?
This happens extremely randomly in Claude Code, on conversations that really at not very long.
Including all tool calls, this conversation is 16528 characters, and now I have to jump through hoops to continue it. This must be the tenth time I've hit this in a month. It would be nice to see a response.
I have run into this issue more often and been blocking almost after every tool call that results in a lengthy output.
Since I’m using directly “mcp_server” parameter with the sdk, I don’t even get an opportunity to truncate the 3rd part tool response before reaching the model.
Given this ticket is more than a month old, I hope there is at least some workaround that community has figured out?
Appreciate help, thanks!
This issue has been reported before, here is a link to that "Issue": https://github.com/anthropics/claude-code/issues/760
There is some advice on how to proceed, but I haven't found it very useful, to be honest. You might be lucky, though.
This issue has been inactive for 30 days. If the issue is still occurring, please comment to let us know. Otherwise, this issue will be automatically closed in 30 days for housekeeping purposes.
This issue is still happening and the recovery suggestions from the other issue are not working for me.
Happened to me for the first time just now while using the CC VS code extension. Commenting here because the root cause of the error is likely identical.
I did nothing besides continue the previous conversation. I haven't hit any limits and tried fixing the issue by
/compact, which was interrupted by the same error, and thenScreenshots attached.
<img width="766" height="682" alt="Image" src="https://github.com/user-attachments/assets/a02686f6-0967-4bec-b7b2-561821f624bb" />
<img width="552" height="501" alt="Image" src="https://github.com/user-attachments/assets/46fc6a49-72f4-48ce-8562-04b9e2af40e2" />
@AsteroidHunter how did you retrieve the conversation history?
@thegrahamking They are stored in
~/.claude/projects/{your repo name}/in.jsonlformat.I asked Opus to create a script that converts the
.jsonlinto a text file with neatly separated user and assistant messages. The file names look like hash codes, so when I want to convert a specific chat into a.txt, I just provide a portion of the chat's title to Opus (any sentence or paragraph from the chat would work), and Opus finds and converts it.Pretty useful for retrieving information as the VSCode extension has many UI bugs!
This is a critical disaster issue. I have encountered it 5 times in 3 days and it almost always causes project corruption that is fatally lossy. It worse makes dumb assumptions that it can GIT recover when it has explicit information that GIT is not available and is NOT the SCM model. This is for the $200 maximum service support system.
Its still happening for me till date, any work around for this ?
This happens too much at this point. I have found workarounds that seem ambiguous at best at what context is brought over to the new chat but this should be easily solvable. an external check on context filling and a hook to trigger compaction even at full.
Have you all updated claude code? I haven't faced this issue since my last complaint earlier.
Claude Code updates itself and this issue affects the web version too.
Affecting me today, still not fixed. Double escape does nothing.
<img width="792" height="731" alt="Image" src="https://github.com/user-attachments/assets/5679b5a1-fdd1-4a76-b6a0-95bb08544bd9" />
<img width="791" height="685" alt="Image" src="https://github.com/user-attachments/assets/9c656e9c-13ca-419e-8f4e-1789c90591af" />
Closing for now — inactive for too long. Please open a new issue if this is still relevant.
This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.