### 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? Claude no longer seems to show it's thinking patterns, or very few of them. Sometimes Claude will sit there for 10 minutes, token usage going up, and I'll

[BUG] Can't see Claude's thinking anymore

Open 💬 19 comments Opened Feb 8, 2026 by nullbio

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?

Claude no longer seems to show it's thinking patterns, or very few of them. Sometimes Claude will sit there for 10 minutes, token usage going up, and I'll have a single message in my history (ctrl+o will show 1 single thought, and nothing in the non-ctrl+o view).

Opus 4.6 has a tendency to get stuck in thinking loops too, so not being able to catch these due to lack of visibility is a massive problem.

Having Claude sit there for 10 minutes without a single response or 0 or 1 thoughts shown is anxiety inducing and makes me feel like something is broken. We lose all feedback so we have no idea whether to wait another 10 minutes or kill the session and try something else. I'm hoping this is a bug and not a design change -- if it is a design change, it's a terrible one.

What Should Happen?

Show Claude's thinking patterns again so we can trace how the model is reasoning, and if its reasoning is making sense and it's on the right path. Without this feedback it's impossible to know whether extended thinking sessions are actually gaining any progress or not, which causes anxiety and provides a terrible user experience.

Error Messages/Logs

Steps to Reproduce

Self explanatory

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.37

Platform

Anthropic API

Operating System

Ubuntu/Debian Linux

Terminal/Shell

WSL (Windows Subsystem for Linux)

Additional Information

_No response_

View original on GitHub ↗

19 Comments

github-actions[bot] · 5 months ago

Found 3 possible duplicate issues:

This issue will be automatically closed as a duplicate in 3 days.

If your issue is a duplicate, please close it and 👍 the existing issue instead
To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

joukevandermaas · 4 months ago

The issues mentioned are not duplicate and this is extremely critical, the app is borderline unusable in its current state as claude is just spinning burning tokens with no way for the user to know if it's on track.

danperignon · 4 months ago

Confirming this regression with forensic evidence from v2.1.45 → v2.1.47.

Auto-update enabled; thinking traces worked normally on 2.1.45 this morning. After auto-update to 2.1.47, thinking blocks stopped entirely. One session spans both versions, pinpointing the transition.

Parsed every session JSONL from today across all projects:

10:21  v2.1.45        94 thinking blocks  ← working
10:38  v2.1.45→2.1.47  3 thinking blocks  ← auto-updated mid-session
10:52  v2.1.47         0 thinking blocks
10:58  v2.1.47         0 thinking blocks
11:44  v2.1.47         0 thinking blocks
...
15:48  v2.1.47         0 thinking blocks

20+ sessions across multiple projects on 2.1.47 — every one has zero thinking or redacted_thinking content blocks in the JSONL. The API isn't returning them. This is not a rendering/display issue; the blocks are absent from the raw conversation data.

Setup: Opus 4.6, /model shows "high" effort, verbose mode (Ctrl+O) enabled, thinking toggled on. All confirmed working on 2.1.45 immediately prior.

Not a duplicate of the bot-suggested issues — those are about thinking blocks corrupting conversations. This is about the client silently stopping to request thinking from the API altogether.

nullbio · 4 months ago

Confirming this regression with forensic evidence from v2.1.45 → v2.1.47. Auto-update enabled; thinking traces worked normally on 2.1.45 this morning. After auto-update to 2.1.47, thinking blocks stopped entirely. One session spans both versions, pinpointing the transition. Parsed every session JSONL from today across all projects: `` 10:21 v2.1.45 94 thinking blocks ← working 10:38 v2.1.45→2.1.47 3 thinking blocks ← auto-updated mid-session 10:52 v2.1.47 0 thinking blocks 10:58 v2.1.47 0 thinking blocks 11:44 v2.1.47 0 thinking blocks ... 15:48 v2.1.47 0 thinking blocks ` 20+ sessions across multiple projects on 2.1.47 — every one has **zero** thinking or redacted_thinking content blocks in the JSONL. The API isn't returning them. This is not a rendering/display issue; the blocks are absent from the raw conversation data. **Setup:** Opus 4.6, /model` shows "high" effort, verbose mode (Ctrl+O) enabled, thinking toggled on. All confirmed working on 2.1.45 immediately prior. Not a duplicate of the bot-suggested issues — those are about thinking blocks _corrupting_ conversations. This is about the client silently stopping to request thinking from the API altogether.

Anthropic really needs to figure out how to stop reintroducing bugs they fix so frequently. It's getting to the point where we can't trust any updates to have stability.

man2machine · 2 months ago

I'm still getting this error myself as well. I recall a couple weeks back I saw all the thinking text, and now (presumable after an update) I don't anymore.

cagostino · 2 months ago

I see thinking only when using non-anthropic models within claude code.

nharrer · 2 months ago

In the Claude Code VS Code extension, Claude's thinking/reasoning is displayed in real time while it works. This is extremely valuable for catching wrong directions early, if Claude is going down the wrong path, you can see it and intervene before it wastes time.
In the CLI, this thinking output is hidden entirely. By the time you see a response or action, Claude may have already committed to a flawed approach or at least burned a lot of token which could have been stopped earlier.

aliyenidede · 2 months ago

In the Claude Code VS Code extension, Claude's thinking/reasoning is displayed in real time while it works. This is extremely valuable for catching wrong directions early, if Claude is going down the wrong path, you can see it and intervene before it wastes time. In the CLI, this thinking output is hidden entirely. By the time you see a response or action, Claude may have already committed to a flawed approach or at least burned a lot of token which could have been stopped earlier.

For almost 2 weeks (maybe even more) you cant also see in vscode extension. What are you talking about?

nharrer · 2 months ago

For almost 2 weeks (maybe even more) you cant also see in vscode extension. What are you talking about?

I just checked. My installed extension is up-to-date: anthropic.claude-code 2.1.126

And thinking is displayed:

You have to click on "Thought for xs" to drop down the thought text.

aliyenidede · 2 months ago

> For almost 2 weeks (maybe even more) you cant also see in vscode extension. What are you talking about? I just checked. My installed extension is up-to-date: anthropic.claude-code 2.1.126 And thinking is displayed: <img alt="Image" width="800" src="https://private-user-images.githubusercontent.com/8344095/586941892-e99cc7d6-12dd-41d3-bab2-e6b4d7765cad.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.0_x1BWbQKe1WXXuufU6218SkSLmlWwwdJsP_IbsLt7E"> You have to click on "Thought for xs" to drop down the thought text.

I have also same version. not have any option for it.

nharrer · 2 months ago

I have also same version. not have any option for it.

I deleted my last comment. I found out that the problem was the selected model. The "Thinking" feature works if Sonnet 4.6 is selected in the VS Code extension, which I use most of the time. When I switch to Opus 4.7, I have the same problem as you.

Looking through other issues, I found this workaround:
https://github.com/anthropics/claude-code/issues/54348#issuecomment-4337318753

Hopefully they fix it in general soon. In many cases, Claude is unusable when the thinking is not displayed. It runs for minutes, burns tokens, and we have no idea what it’s doing.

ashishgautam0 · 1 month ago

Hey anthropic you were like safe ai and auditable ai and constitution now what happened

MarkProminic · 1 month ago

Still an issue . . . I have found claude many many times going off track and if I don't correct its thoughts it will implement and has implemetned things I never asked for, which I don't want or need.

REtoolsx · 28 days ago

Still reproduces on v2.1.181 (latest), Windows 11 / Windows Terminal.

A data point that differs from the discussion above: thinking is hidden on both Opus AND Sonnet for me. Switching to Sonnet 4.6 does NOT restore it (several comments here suggest this is Opus-only — so for me it looks model-independent). Extended thinking is enabled and the spinner + token counter run, but no thinking is shown inline, and Ctrl+O (transcript) followed by Ctrl+E (show all content) does not reveal it either.

So: still present on the latest release, and not limited to Opus. +1

constantinosbou · 23 days ago

We still have this bug and its extremely annoying. I literally feel like am getting robed, the model is thinking after every message for way to long, im just sitting on my computer counting pixels. wtaf

duaneking · 19 days ago

This is an insane bug that needs to be fixed, This is insane. I must know what the thing is thinking because it's spiraling out of control and it looks like it's doing nothing, and then it does something random that's hallucinated and I'm not able to recover from it because I didn't know it was coming, because I wasn't allowed to monitor it.

cagostino · 19 days ago

at this point just use any other model in claude code, the thinking shows for all ollama ones and its the only way this product is tolerable. they've done some voodoo around when claude models think to prevent ppl from extracting the thought traces

Artem-Kuroptev-Upper · 13 days ago

Solution from [https://hypogray.com/stories/claude-code-hides-thinking](url) helped me. Set "showThinkingSummaries": true in ~/.claude/settings.json, restart app/extension and you (probably) will be able to check thinking process

nharrer · 13 days ago

Thanks! I can confirm setting "showThinkingSummaries": true worked. Tested it in claude cli and the vscode extension: