/context command fails with LiteLLM proxy (Bedrock)

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?

The /context slash command fails with error:
Error: 500 - max_tokens must be greater than thinking.budget_tokens

Occurs even with alwaysThinkingEnabled: false in settings.json

What Should Happen?

Command should respect alwaysThinkingEnabled setting or configure tokens properly

Error Messages/Logs

Error: 500 {"error":{"message":"Error calling litellm.acompletion for non-Anthropic model: litellm.BadRequestError:
  BedrockException - {\"message\":\"The model returned the following errors: `max_tokens` must be greater than
  `thinking.budget_tokens`. Please consult our documentation at https://docs.claude.com/en/docs/build-with-claude/extended-thinking#
  max-tokens-and-context-window-size\"}","type":"None","param":"None","code":"500"}}

Steps to Reproduce

1. Configure LiteLLM proxy to Bedrock
2. Set alwaysThinkingEnabled: false in settings.json
3. Run /context
4. See error

Tried DISABLE_INTERLEAVED_THINKING=true - didn't help

Claude Model

Not sure / Multiple models

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.0.28

Platform

AWS Bedrock

Operating System

macOS

Terminal/Shell

iTerm2

Additional Information

Claude Code generated this more detailed bug report when prompted:

Bug Report: /context Command Fails with Bedrock Models via LiteLLM Proxy

Summary:
The /context slash command consistently fails with a max_tokens vs thinking.budget_tokens error when using Claude models through a
LiteLLM proxy (Bedrock backend), even when extended thinking is explicitly disabled.

Environment:

• Claude Code Version: 2.0.28
• Model: claude-sonnet-4-5 (via LiteLLM proxy to AWS Bedrock)
• OS: macOS (Darwin 25.0.0)
• Configuration: Using LiteLLM proxy with custom ANTHROPIC_BASE_URL

Settings Configuration:

  {
    "env": {
      "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-sonnet-4-5",
      "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-5",
      "ANTHROPIC_DEFAULT_HAIKU_MODEL": "claude-haiku-4-5-20251001",
      "CLAUDE_TRACE_SUBAGENTS": "1"
    },
    "alwaysThinkingEnabled": false
  }

Error:

  Error: 500 {"error":{"message":"Error calling litellm.acompletion for non-Anthropic model: litellm.BadRequestError:
  BedrockException - {\"message\":\"The model returned the following errors: `max_tokens` must be greater than
  `thinking.budget_tokens`. Please consult our documentation at https://docs.claude.com/en/docs/build-with-claude/extended-thinking#
  max-tokens-and-context-window-size\"}","type":"None","param":"None","code":"500"}}

Steps to Reproduce:

1. Configure Claude Code with LiteLLM proxy pointing to Bedrock backend
2. Set alwaysThinkingEnabled: false in settings.json
3. Run /context command
4. Observe error

Attempted Workarounds (None Successful):

• /context thinking mode off
• /context extended thinking disabled
• export DISABLE_INTERLEAVED_THINKING=true followed by /context

All attempts resulted in the same error, indicating the /context command may have extended thinking parameters hardcoded that
cannot be overridden.

Expected Behavior:
The /context command should either:

1. Respect the alwaysThinkingEnabled: false setting and not use extended thinking, OR
2. Configure max_tokens high enough to accommodate the thinking.budget_tokens it's requesting

Actual Behavior:
The command attempts to use extended thinking with incompatible token budget configuration, causing Bedrock to reject the request.

Impact:
Users cannot use the /context command to check context window usage when using Bedrock models through LiteLLM proxy.

Notes:

• Regular conversations work fine with the same configuration
• Other slash commands work without issues
• This appears specific to the /context command's implementation when used with non-native Anthropic API endpoints