[BUG] Thinking mode toggle with tab button isn't working in 2.0.67

[afaraz009]

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?

It was very convinient to switch the thinking mode with a shortcut key like Tab. Now in the latest release I have to go to the /config and toggle thinking to enable or disable it

What Should Happen?

There should be a shortcut way to toggle thinking mode

Error Messages/Logs

Steps to Reproduce

.

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

No response

Claude Code Version

2.0.67

Platform

Anthropic API

Operating System

Windows

Terminal/Shell

PowerShell

Additional Information

No response

[bchaipats]

It sucks I know. But, try /config to switch Thinking off while waiting for them to bring this toggle key back.

At least "Thinking Off" should be default, it is more sensible IMO.

reference: https://github.com/anthropics/claude-code/commit/2192c86c208b39aa262ada534d89f72f65708a8e

[rbarcante]

That sucks, should at least introduce a new shortcut for that

[gajop-ptx]

Indeed, it's quite annoying to have to go to /config since it takes a bit of time to navigate and you can't switch to that mid-work, so you have to interrupt the AI first.

[bis0908]

Please bring this feature back!!

[ikemo3]

I keep Thinking mode disabled by default due to the 5-hour thinking budget limit. When I'm not satisfied with Claude's output, I turn it on to get better reasoning. Typing /config adds 7 unnecessary keystrokes per switch.

But the real blocker is that I can't toggle Thinking or change models while editing. I have to either plan ahead or discard what I'm typing. Same issue with model switching.

Claude is capable enough that neither Thinking nor Opus are needed for every task. For simple tasks that need speed, I use Haiku. The point of shortcuts is to enable dynamic switching and optimize token usage. Without them, the optimization doesn't work. (Of course, it'd be ideal if Thinking and Opus's intelligence, and Haiku's speed were all cheap enough to use freely.)

[afaraz009]

Same issue with model switching.

You can use Alt+p to switch models while editing

[aristeoibarra]

+1. Please add an alternative shortcut for thinking mode toggle. Going through /config breaks the workflow.

[kvasanova]

I also found this problem today. Will have to return it. Ideally they have to make customizable shortcuts for popular(or any) action

[rrcobb]

I accidentally hit this Tab toggle all the time, so the motivation for the removal is understandable. However, the setting for the default isn't actually respected! I have to go to /config every time I start a conversation...

Also, an alt keyboard shortcut for this would be nice!

[iBeej]

Discovered this problem today, because Claude kept thinking and chewing up more tokens. Couldn't turn it off with TAB. Discovered it's in /config but it's more annoying to get to AND it doesn't persist across sessions. Will switch back to enabled on session start. I really loved the TAB shortcut for thinking mode. I hope this is a bug and not a "feature" to remove it.

[LuisFGomez]

No! Muh tab!

Why break existing <TAB> behavior? Couldn't you just have used Ctrl+Space for prompt completion? It at least has history of being an "auto complete" shortcut.

[bogini]

Thanks for the reports. We removed the Tab shortcut because it was causing frequent accidental toggles. In our testing, thinking mode both reduces overall token usage and improves results, so we'd recommend keeping it on for most workflows.

We're aware the settings persistence is broken, a fix is coming in the next release.

[Letiliel]

Thanks for the reports. We removed the Tab shortcut because it was causing frequent accidental toggles. In our testing, thinking mode both reduces overall token usage and improves results, so we'd recommend keeping it on for most workflows.

We're aware the settings persistence is broken, a fix is coming in the next release.

Thank you for replying. Tab wasn't the right shortcut indeed. The solution to accidental toggles was to change it or allow to configure it though, not remove it entirely.

I personally find that thinking mode slows down Claude a lot, and in frequent cases seems to make it ramble incoherently with chains of "oh but wait" contradictions. Also want to disable it when I'm reaching end of context window and need quick completion. It's a case by case tool for me.

The statement that thinking reduces token usage is interesting. User's should be informed clearly if that's the case, with supporting data, because it's not intuitive.

[Anthony-Bible]

i agree, half the time i'm working on a simple script that doesn't need thinking at all and jut makes it take longer than needed so being able to quickly toggle off and on is critical. I don't want to exit and change it every time. Even a slash command (like /model) isn't as user friendly IMO.

[certainmagic]

Given the issues you were seeing, switching the keypress for toggling thinking mode seems reasonable (although it could've been made more obvious that it had changed).

I've been toggling thinking mode quite often because it seems to use my quota rapidly. I reserve it for when I'm in planning mode on something non-trivial and turn it off to execute the plan. I like to see Claude's reasoning. Sometimes the thinking output mentions things I hadn't noticed and which it doesn't think are important, but which turn out to be useful. That can help me guide it when it's not thinking about something the right way or help me realize there's a better way to get something done.

That said, it would be nice to have fewer controls, especially as Opus 4.5's level-of-effort setting becomes available. There are getting to be several controls whose interactions aren't obvious. I don't envy you having to find a way to explain the various controls and choose which to make easily accessible. In the meantime, it'd be nice to have a shortcut or /command to see and change the level without having to use the interactive menu.

thanks!

[zxzinn]

You can add this to ~/.claude/settings.json:

{
  "env": {
    "MAX_THINKING_TOKENS": "0"
  }
}

This disables thinking mode completely even toggled in /config, but this is a temporary workaround for people who don't want thinking by default, if you want to toggle thinking again you need to remove this configuration.

[dulev]

Thanks for the reports. We removed the Tab shortcut because it was causing frequent accidental toggles. In our testing, thinking mode both reduces overall token usage and improves results, so we'd recommend keeping it on for most workflows.

We're aware the settings persistence is broken, a fix is coming in the next release.

Will there be a way to set our own shortcut for this? For trivial changes it really degrades the DX to have to use thinking mode.

[Elijas]

Just run claude with MAX_THINKING_TOKENS=0 claude or create your own wrapper for it fast-claude and add it to your PATH so you can launch fast-claude from anywhere

[ipeterov]

I use Claude Code a lot, and I always kept thinking turned off - I try to think together with the model, throwing quick messaages back and forth. Having thinking on breaks that completely - Claude keeps starting internal monologues that sometimes last a minute, and I have to sit around and wait for it to produce a small, and maybe already irrelevant answer.

Maybe thinking is good if you're trying to get better one-shot responses, but that's not what Claude is about for me. I constantly steer, adjust, and iterate, and that doesn't work now. (the fact that the setting keeps resetting to thinking: on, doesn't help)

[dhisonp]

I'd be keen to see the implied data of thinking mode saving token usage in general. Perhaps in high-level or decision-intensive tasks, but simple code it's really tedious to be forced into Thinking mode without an easy way out i.e. shortcut.

But yes, thanks for the reply!

[Elijas]

I'd be keen to see the implied data of thinking mode saving token usage in general. Perhaps in high-level or decision-intensive tasks, but simple code it's really tedious to be forced into Thinking mode without an easy way out i.e. shortcut.

But yes, thanks for the reply!

I second that

shortcut doesnt have to be a [TAB] Some cmd + t or a sequence like cmd + / -> t would be arguably just as ergonomic

[NasAndNora]

This was working perfectly in the previous version - the Tab key toggle was simple and intuitive. The new update broke this functionality.

Beyond this issue, Claude Code has multiple bugs right now, and Opus as a model has gone from incredible to catastrophic in terms of quality.

I really hope the team addresses these issues soon.

[andreaganduglia]

We removed the Tab shortcut because it was causing frequent accidental toggles.

@bogini

IMHO, the correct solution to avoid accidental toggles is to improve the shortcut, not remove it entirely, forcing users to go through the settings to enable or disable the feature.

[mimkorn]

the hell — i never accidentally toggled it and don't understand why it should ever happen. Also, the feedback in CLI is so clear and any accidental toggle is so easy to revert with another tap on the tab that it's a non-issue. Who is ever going to be going through config to turn it off and on in daily flow.

[fadedlamp42]

I had reported this myself when I noticed it and, after another case recently where the mobile app had thinking set to on by default, I decided for myself that there's something more profit-driven at heart here (which is a real shame given Anthropic's stated mission).

Start here if what I'm saying seems to resonate, I'm keeping my own issue open for meta-discussion of potential deeper problems here.

I'd be keen to see the implied data of thinking mode saving token usage in general. Perhaps in high-level or decision-intensive tasks, but simple code it's really tedious to be forced into Thinking mode without an easy way out i.e. shortcut.

Thirded, this data should have been published and cited within the first reply IMO (or at least taken this issue as a way to quietly publish this if they didn't want to draw so much attention to it, but if it were true, I think they'd be making a big PR push to get everyone to turn thinking mode on themselves; I just think they might be worried the average user doesn't even know what thinking is and won't notice it getting set on or off/consuming their limits more quickly).

[niksauer]

Can we please get a shortcut similar to alt + p (for switching models)? 😕

[torlando-tech]

this bug seems to be back in 2.0.70

[aristeoibarra]

In version 2.0.70, the Thinking mode indicator is no longer visible in the input field. It should be displayed explicitly so users know when Thinking mode is active.

[gajop-ptx]

I'm struggling to make use of the current claude code due to this, especially since I rely on Opus almost exclusively.

My workaround is to keep using the older version:

npm install -g @anthropic-ai/[email protected]

& disable auto-updates in ~/.claude/settings.json

{
   "env": {
      "DISABLE_AUTOUPDATER": "1"
   }
}

Hopefully they reintroduce some shortcut so I don't have to use this older version.

[SebasScript]

+1 one to bring this back, Caude is responding faster with no thinking which is highly desired for certain tasks, questions, simple queries, having to use config or running a separate instance only for this degrades the UX