Issue Station
Sign In
Tools hereToolbox

claude-code - 💡(How to fix) Fix [FEATURE] Allow forcing extended thinking on Opus 4.7 [1 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
anthropics/claude-code#49555•Fetched 2026-04-17 08:37:49
View on GitHub
Comments
0
Participants
1
Timeline
4
Reactions
0
Author
cminn10cminn10
Participants
cminn10cminn10
Timeline (top)
labeled ×2cross-referenced ×1renamed ×1

Per the docs, Opus 4.7 always uses adaptive reasoning, and CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING / MAX_THINKING_TOKENS no longer apply. Please provide a way to opt out of adaptive reasoning on 4.7, the same way 4.6 allows.

Root Cause

Per the docs, Opus 4.7 always uses adaptive reasoning, and CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING / MAX_THINKING_TOKENS no longer apply. Please provide a way to opt out of adaptive reasoning on 4.7, the same way 4.6 allows.

RAW_BUFFERClick to expand / collapse

Summary

Per the docs, Opus 4.7 always uses adaptive reasoning, and CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING / MAX_THINKING_TOKENS no longer apply. Please provide a way to opt out of adaptive reasoning on 4.7, the same way 4.6 allows.

Motivation

Adaptive reasoning under-thinks on questions where the premise is the trap, not the surface complexity. The model decides "this looks routine" before loading enough context to notice the routine answer is wrong.

Example from today:

Me: "I want to wash my car. The car wash is 100m away. Should I walk or drive?"

Opus 4.7 (adaptive): "Walk. 100m is a 1-2 minute walk..."

Me: "OMG, I'm going to wash my car, if I don't drive, what am I going to wash?"

Opus 4.6 with thinking enabled reportedly gets this on the first try. This generalizes: code review, debugging, and architectural trade-off questions all suffer when the model commits before reasoning about whether it's solving the right problem. On 4.6 I could mitigate with CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING=1 + MAX_THINKING_TOKENS. On 4.7 there's no escape hatch.

Proposed Solutions

  1. Re-enable CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING for Opus 4.7, gated by MAX_THINKING_TOKENS — keeps a single, unified config across all models.
  2. A /think slash command or prefix that forces extended thinking on the next turn, for cases where the user knows the question is subtler than it looks.

Why prompt-based "think harder" isn't enough

The docs say the model responds to such guidance within its effort setting. That caveat is the issue — for the failure mode above, the model has already decided the prompt doesn't warrant thinking before it weighs the instruction. A user-controlled override sidesteps that.

Environment

  • Claude Code version: 2.1.112
  • Model: claude-opus-4-7[1m]
  • OS: macOS 25.3.0

extent analysis

TL;DR

Re-enabling CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING for Opus 4.7 or introducing a user-controlled override like a /think slash command could address the issue of adaptive reasoning under-thinking on subtle questions.

Guidance

  • Consider re-enabling CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING for Opus 4.7 to provide a unified config across models, allowing users to opt out of adaptive reasoning when needed.
  • Introducing a /think slash command or prefix could offer a flexible solution, enabling users to force extended thinking on specific questions that require more in-depth analysis.
  • Evaluate the effectiveness of these solutions in scenarios where the premise is the trap, rather than the surface complexity, to ensure they adequately address the under-thinking issue.
  • Assess the potential impact of these changes on the model's performance and response time, as increased thinking tokens or disabled adaptive reasoning might affect the overall user experience.

Example

No code snippet is provided as the issue focuses on configuration and model behavior rather than specific code implementation.

Notes

The proposed solutions aim to address the specific failure mode described, where adaptive reasoning under-thinks on questions with subtle premises. However, the effectiveness of these solutions may vary depending on the complexity and nature of the questions being asked.

Recommendation

Apply a workaround, such as re-enabling CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING or introducing a /think command, as these solutions directly address the reported issue and provide a means for users to control the model's thinking behavior when necessary.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…
#api#ssr#database connection#vector store#embedding generation

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

claude-code - 💡(How to fix) Fix [FEATURE] Allow forcing extended thinking on Opus 4.7 [1 participants]