TL;DR
Claude Code is reportedly returning blank thinking blocks even though the associated reasoning still counts toward usage costs. The report raises billing-transparency concerns, but the cause, scope, affected versions and Anthropic’s response remain unconfirmed.
Claude Code is reportedly displaying blank thinking blocks while the underlying reasoning continues to count toward usage costs, according to a report from The Register. The behavior matters because users may pay for model activity without receiving the visible reasoning output they expected.
The published report identifies two linked behaviors: a thinking block appears blank, while the reasoning associated with that response remains billable. The headline does not provide a token count, price estimate, reproduction procedure or explanation for why the visible block contains no text.
A blank block does not, on its own, establish that no reasoning occurred. It indicates that the interface returned no visible content in that part of the response. The continued charge suggests that computation or reasoning tokens may still have been processed, although the available details do not establish how those tokens were measured.
The report does not say whether the behavior affects all Claude Code users, a limited group of accounts, particular models or a specific release. It also does not establish whether users encounter the problem consistently or only under certain prompts, tools, environments or output settings.
Blank thinking blocks. Reasoning costs remain.
Claude Code is reportedly showing empty thinking blocks while the reasoning associated with those responses still counts toward usage costs. The report raises a transparency question—not proof of incorrect billing.
No reasoning text appears inside the reported thinking block.
Reasoning activity reportedly continues to register as usage.
Cause, scope, affected releases, fixes and credits remain unclear.
Facts, implications and open questions
The narrow finding concerns the gap between what customers can see and what usage records may count. Several broader explanations remain possible.
A block appears empty
The interface reportedly returns no visible reasoning content in the affected thinking block.
This is a presentation observationCosts are harder to trace
Developers have less visible information for connecting recorded reasoning usage to the response they received.
This creates a transparency gapWhich layer failed?
The available report does not distinguish among rendering, API response, model policy or metering behavior.
This requires technical evidenceOne request, four separate layers
A blank block does not prove that no computation occurred. Internal reasoning, returned fields, interface rendering and billing records are related—but they are not identical.
Model computes
Internal processing may occur without exposing a complete chain of thought.
Response forms
The system packages output fields, metadata and usage information.
Client renders
Claude Code decides what appears in the visible thinking block.
Usage records
Metering may count reasoning activity independently of displayed text.
The crucial distinction
Visible reasoning text is not necessarily a direct transcript of every internal or billable token. The concern is that apparently empty output can prevent customers from understanding why reasoning usage was recorded.
What is known—and what is not
The report supports two limited claims. It does not yet support conclusions about incorrect metering, universal impact or remediation.
| Question | Current reading | Status | What would confirm it? |
|---|---|---|---|
| Does the thinking block appear blank? | Reportedly yes in affected responses. | ✓Reported | Reproduction steps, screenshots and request records |
| Does reasoning still count toward usage? | The report says associated reasoning remains billable. | ✓Reported | Itemized usage logs linked to specific requests |
| Does blank output mean no reasoning occurred? | No. Absence of visible text does not prove absence of computation. | ✗Not established | Internal telemetry or an official technical explanation |
| Are the charges incorrect? | The available information cannot distinguish valid metering from a billing fault. | ~Unconfirmed | Metering documentation, logs and Anthropic review |
| Are all users or versions affected? | No affected models, account types or client releases are identified. | ~Unknown | Incident scope, version matrix and prevalence data |
| Has a fix or refund been announced? | No patch, credit program or pricing change is confirmed. | ✗Not confirmed | Official advisory, release note or billing notice |
Cost impact cannot yet be sized
No token count, price estimate, frequency data or affected-user total has been provided. The financial effect could be minor or material; the available report does not determine which.
Preserve the trail when a blank block appears
Blank Output Complicates Cost Tracking
The issue creates a practical problem for developers who monitor token consumption and evaluate whether an assistant’s output justifies its cost. When a billed reasoning segment contains no visible text, users have less information for connecting usage charges to the response they received.
That gap may also make troubleshooting harder. Developers cannot readily tell whether a blank block reflects an interface-rendering problem, intentionally hidden reasoning, an empty response field or another failure. Each explanation has different implications for reliability, billing and whether work needs to be repeated.
The report does not show that the charge itself is incorrect. Reasoning models can perform internal computation without exposing a full chain of thought, and visible output is not necessarily a direct record of every billed token. The narrower concern is that apparently empty output may leave customers unable to understand why reasoning usage was recorded.
OBD2 Scanner, MUCAR 632 Elite AI-Assisted Bidirectional Scan Tool, 15 Reset Services Oil/TPMS/EPB/BMS/SAS/Brake/Throttle Car Scanner Diagnostic Tool, AutoAuth FCA, CANFD, AutoVIN, Lifetime Free Update
- Lifetime Free Updates: No subscription fees, lifelong updates included
- AI-Assisted Troubleshooting: Instant AI analysis and repair suggestions
- Comprehensive Code Reading: Reads and erases engine, ABS, SRS, transmission codes
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Claude Code’s Reasoning Visibility
Claude Code is Anthropic’s coding assistant for software-development tasks. Thinking or reasoning indicators can help users understand when the model is processing a request, but the displayed material may not correspond exactly to the system’s internal computation or billing records.
The reported behavior sits at the intersection of product presentation and usage accounting. A display fault could leave valid metering intact, while a metering problem could require billing corrections. The headline alone does not identify which layer is responsible.
No technical advisory, incident report or detailed statement from Anthropic was included in the available information. There is also no supplied evidence of refunds, account credits, a software patch or a change to Claude Code’s pricing policies.
AI model billing transparency tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Cause and Scope Stay Unconfirmed
It is not yet clear whether the blank blocks result from a Claude Code interface defect, an API response issue, a model-output policy or expected behavior that is being displayed poorly. The report also leaves the affected model versions, account types and client releases unidentified.
The size of the cost impact remains unknown. No data has been provided on how many users are affected, how often the behavior occurs or how many billable tokens a blank block may represent. Without itemized examples or usage logs, readers cannot determine whether the charges are minor or material.
Anthropic’s position is also unclear from the information provided. There is no confirmed explanation of whether the company recognizes the behavior as a bug, considers the billing accurate or plans to offer credits. Claims beyond the reported blank output and continued cost would be premature.
AI development environment monitors
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Users Await Billing Clarification
The next meaningful development would be a technical explanation from Anthropic identifying whether the issue lies in display, model output or usage metering. Users will also need details on affected releases, any available workaround and whether past charges will be reviewed.
Until more information is published, customers can compare blank responses with their own usage records and preserve timestamps, model names and request identifiers when reporting the behavior. Those records could help distinguish an isolated rendering fault from a broader billing-visibility issue.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is happening in Claude Code?
Claude Code reportedly returns a blank thinking block while the reasoning tied to that response still counts toward usage costs. Further technical details have not been confirmed.
Does a blank block mean Claude performed no reasoning?
No. A blank display shows only that no reasoning text was visible in that block. It does not prove that no internal computation took place.
Are the reported charges incorrect?
That has not been established. The report says reasoning remains billable, but it does not provide enough evidence to determine whether the metering is wrong, the interface is faulty or both systems are behaving as designed.
Has Anthropic announced a fix or refunds?
No fix, refund program or billing adjustment is confirmed in the information available. Users are waiting for an official explanation covering the cause, scope and treatment of affected charges.
Source: Anthropic
Source: Anthropic
