Why is extended thinking greyed out in Claude?

The most common reasons are that the model you are using does not support extended thinking, or your plan does not include access. Switch to a supported model in the claude.ai interface and check your plan on the Anthropic account page.

Why is extended thinking not showing a reasoning block?

Check that the feature toggle is on, that the interface is not hiding the block by default, and that the budget_tokens value in the API is large enough if you are using the API. Some interfaces collapse the reasoning block; look for an expand control.

Does extended thinking work with Claude Code?

Claude Code triggers extended thinking automatically on harder tasks. You do not always control whether it activates. If you need it on a specific task, phrase the request in a way that makes the complexity clear, or check the Claude Code documentation for explicit configuration options.

What error do I get if extended thinking is misconfigured in the API?

The API returns a validation error if temperature is not set to 1, or if the model does not support thinking. Check the response body for the specific error message and compare it against Anthropic's current API reference.

Is the thinking block always visible?

The model generates a reasoning chain, but the interface can render it collapsed or not show it prominently. Look for a thinking indicator or collapse control in the response area. In the API, the thinking text comes in a separate content block from the final response.

Claude Extended Thinking Not Working? Checks and Fixes

Extended thinking not appearing in Claude? The cause is usually model access, plan limits, or a configuration detail. This guide separates each cause and gives a concrete fix.

October 8, 2026Updated October 21, 20265 min read

Claude extended thinking stops working for a predictable set of reasons: the model does not support it, the plan does not include access, the feature is not enabled for the session, or the API request has a configuration error. Identifying which applies to your situation takes under five minutes.

If the thinking block is not appearing when you expect it, start with the model and plan check before changing anything else. While investigating, tools like fenbs let you save the problem description and checkpoint so the work is not lost if the session has to restart. Extended thinking is covered in the Claude extended thinking guide if you want to understand the feature first.

Check 1: model access

Extended thinking is only available on models that support it. In the Claude web app, it is linked to specific model versions. If you are using an older model or a model that does not include extended thinking, the control will not appear.

Fix: in the claude.ai interface, check which model is selected. If extended thinking is not visible, switch to a model that supports it. Anthropic's model documentation on their site lists which models include extended thinking.

Check 2: plan or account access

Extended thinking may require a paid plan or a specific plan tier. The free plan often limits access to less capable models, and the thinking controls may not appear.

Fix: check your current plan on the Anthropic account page. If extended thinking is not included, you will need to upgrade or switch to the API if you have a paid API key. Do not assume that having a paid account means all features are unlocked — check the specific plan page for what is included.

Check 3: the feature toggle in claude.ai

In the claude.ai web interface, extended thinking is not always on by default. Look for a toggle or a mode indicator near the message input. It may be labeled "Extended thinking", "Think", or shown as a brain icon, depending on the current interface version.

Fix: if the toggle is available but off, enable it. If it is not visible at all, the cause is likely model or plan access (checks 1 and 2 above).

How fenbs helps

Preserve the problem brief while you investigate

When extended thinking is not working, it is easy to lose the detailed problem brief you assembled. Save it in fenbs as a task or context note before experimenting with different settings. That context stays available to the next session, whether you fix the issue or switch to a different tool while you wait.

Create a free account

Free for one person · No credit card · Works with Claude, ChatGPT, Cursor and more

Check 4: API configuration errors

If you use the Anthropic API directly, extended thinking requires specific parameters in the request. The two most common errors are sending the request to a model that does not support thinking, and setting budget_tokens too low for the problem.

Model mismatch: confirm the model ID in your request matches a model that supports extended thinking. Anthropic's API reference lists supported models.
Budget too low: the budget_tokens value sets a maximum. If it is set to a very small number, the model may not produce a visible thinking block. Try increasing it.
Temperature conflict: extended thinking requires temperature to be set to 1. Setting a different value causes an API error.
Streaming not configured: if you are using streaming, the thinking delta type must be handled in your parser. An unhandled thinking delta can cause it to appear invisible in the output.

Check 5: Claude Code and IDE integrations

In Claude Code, extended thinking is triggered automatically on harder tasks. You cannot always control whether it runs; the model decides. If you expect to see a thinking block on a specific task and do not, the model may have determined the task does not warrant it.

For IDE extensions that connect to Claude, extended thinking availability depends on the model the extension uses and the API version it calls. Check the extension's current documentation. The Claude Code not working guide covers Claude Code-specific issues.

Check 6: hidden reasoning

Some interfaces render the thinking block as collapsed or hidden by default. The model produced a thinking chain, but the UI did not show it prominently. Look for an expand control or a "Thinking..." indicator in the response.

Check 7: request stalling

Extended thinking requests take longer than standard requests because the model generates a reasoning chain first. A request that appears stuck may just be working through a long reasoning process. Wait longer than you normally would before concluding that something is broken. For the API, the timeout on your HTTP client may need to be increased to accommodate long reasoning chains.

If the request times out, reduce the problem scope or break it into smaller questions. A very large budget and a long conversation history can extend generation time significantly. As a rule of thumb, start with a budget of around 10,000 tokens for a medium-complexity problem and adjust from there.

How to do this with fenbs

Create a free account and add the investigation as a task.
Record the problem brief — what you were trying to accomplish and the exact symptoms — in the task's context.
Work through the checks above and record what you find in the task. This creates a log you can hand off to support if needed.
Connect Claude to fenbs from /integrations once the issue is resolved. The next session reads the same brief without restatement.
Save progress before you stop so the checkpoint is clear for the next session.