### Preflight Checklist - [x] I have searched [existing issues](https://github.com/anthropics/claude-code/issues?q=is%3Aissue%20state%3Aopen%20label%3Abug) 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? Setting CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 as a shell environment variable causes all subagents that use Haiku 4.5 (e.g., Explore agent) to fail with an A

[BUG] CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 bypasses model capability checks, breaks Haiku subagents

Status Fixed / completed

Reported on v2.1.104

Maintainer reply ✓ Yes — claude[bot]

Activity 5 comments · opened Apr 13, 2026 · closed May 28, 2026

💡 Likely answer: A maintainer (claude[bot], contributor) responded on this thread — see the highlighted reply below.

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?

Setting CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 as a shell environment variable causes all subagents that use Haiku 4.5 (e.g., Explore agent) to fail with an API error indicating the effort parameter is not supported for the current model configuration.

The env var appears to force the effort parameter onto every API call unconditionally, including calls to models that don't support it. The model-specific capability check that would normally prevent sending effort to Haiku is bypassed entirely when this env var is set.

What Should Happen?

The env var should enable effort only for models that support it (Opus 4.6, Sonnet 4.6, etc.), not override model capability checks. The model guard should take precedence so that Haiku subagents never receive the effort parameter regardless of env var state.

Error Messages/Logs

API Error: 400 {"type":"error","error":{"type":"invalid_request_error","message":"This model does not support the effort parameter."},"request_id":"req_011Ca11Jz7wdaZH9TaWS5sSw"}

Steps to Reproduce

Set the env var: export CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1
Start claude
Trigger any subagent that uses Haiku (Explore agent is the easiest)
Subagent fails: 0 tool uses, 0 tokens, 0s. API rejects the effort parameter.
Unset the env var: unset CLAUDE_CODE_ALWAYS_ENABLE_EFFORT
Restart claude, trigger same subagent. Works normally.

Claude Model

Opus

Is this a regression?

Claude Code Version

2.1.104

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

Other

Additional Information

Other effort-related env vars (MAX_THINKING_TOKENS, CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING, CLAUDE_CODE_EFFORT_LEVEL) do not cause this issue. Isolated through one-by-one elimination testing. Only CLAUDE_CODE_ALWAYS_ENABLE_EFFORT triggers the failure.

Related: #30795 (same error when effort was sent to an unsupported model on Bedrock, fixed after 2.1.68)

View original on GitHub ↗

5 Comments

pitzegluck · 4 months ago

Tested directly against the older binary with CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 set. Same error. This is not a regression - the env var has always bypassed model capability checks. Updated regression status to "No."

marcoabreu · 4 months ago

This is really frustrating because forcing effort not only breaks Explore agents but Claude instead then opts to do the exploration in the main context :(

Leo-Atienza · 3 months ago

Confirming the same root-cause on a different setup — adding repro weight.

Environment

Claude Code 2.1.140 (also repro'd on 2.1.138 from a prior session)
Windows 11, Git Bash shell
Main session model: claude-opus-4-7
Affected subagent: built-in Explore (Haiku 4.5 backed)

Trigger: CLAUDE_CODE_ALWAYS_ENABLE_EFFORT=1 set in ~/.bashrc AND Windows User env (originally added per #13532 to fix the extended-thinking regression).

Repro (deterministic):

Agent({ subagent_type: "Explore", description: "...", prompt: "..." })
→ API Error: 400 This model does not support the effort parameter.

Every Explore call dies before invocation. Claude often surfaces this to users as "Explore agent unavailable," which obscures the real cause.

Workaround applied: unset CLAUDE_CODE_ALWAYS_ENABLE_EFFORT (and the now-inert CLAUDE_CODE_EFFORT_LEVEL) at both shell-profile and Windows-User-env layers. Explore works again immediately.

Trade-off this forces on power users: those of us who set the env var per #13532 to force max effort on the main thread are now choosing between (a) deeper main-thread thinking but no Haiku subagents, or (b) working subagents but losing forced-max effort. The fix proposed in this issue (model-capability guard takes precedence over the env var) would resolve both — eager for it to ship.

claude[bot] contributor · 3 months ago

This issue was fixed as of version 2.1.154.

github-actions[bot] · 1 month ago

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.