Claude Code Returns Empty Thinking Blocks
Anthropic's AI model Claude faces issues displaying its reasoning processes.
1Claude Code Fails to Display Thinking Process
Anthropic's Claude Code API is returning empty thinking blocks for Opus 4.8 and Sonnet 5 models, even when users explicitly request them. The 'thinking' feature allows models to use additional tokens for reasoning through complex problems. Normally, this process provides a summary explaining how the model reached its conclusions. However, these summaries currently fail to display properly, frustrating developers who rely on this transparency.
2Developer Complaints Surface
Developer Michael Hood observed on July 16, 2026, that the API returns empty thinking blocks even with the 'summarized' parameter explicitly requested. The issue is reportedly not widespread and is under investigation. Similar problems have been reported with the Claude Code extension for VS Code. Developers are being charged for thinking processes they can't view, raising concerns about billing transparency. This comes as AI explainability becomes increasingly important for developer trust.
3Token Billing Controversy
A bug report claims thinking summaries are being truncated while token bills remain unchanged. An anonymous developer states, 'Thinking is fully generated and billed, but part of the summary stream is silently dropped.' Anthropic's documentation clarifies users are charged for all thinking tokens generated, even when collapsed or redacted. This policy creates uncertainty about whether bug-related truncations affect billing, highlighting the need for clearer communication about AI service charges.
4API's Operational Logic
Anthropic explains that thinking has a cost: tokens Claude spends reasoning are billed as output tokens, even when the thinking text isn't returned. These tokens count toward max_tokens alongside response text. The company advises customers wanting to reduce thinking costs to lower their budget settings or disable the feature entirely. This billing approach reflects the computational resources required for AI reasoning, regardless of whether users see the intermediate steps.
5Stream Termination During Long Requests
The Anthropic API has been observed terminating data streams during extended thinking sessions. Developer Hector Bernstorff identified this as a client-side issue related to network behavior tuning. The problem involves balancing perceived latency against the risk of stuck requests. Work continues to optimize how the system handles prolonged reasoning tasks, which are computationally intensive but valuable for complex problem-solving scenarios.
6Value of Community Feedback
An Anthropic spokesperson stated, 'Claude Code ships updates nearly daily, and community reports like these GitHub issues are crucial for quickly identifying problems.' The company expressed gratitude to developers filing reports and committed to ongoing fixes. This incident demonstrates how open developer communities contribute to AI tool refinement, with user feedback driving rapid iterations in fast-evolving generative AI technologies.
7Impact on Developer Ecosystem
These thinking process issues raise concerns about trust in AI-assisted development tools. Developers rely on reasoning summaries to understand model outputs, especially for critical applications. The lack of transparency challenges debugging efforts and output validation. As AI becomes more integrated into development workflows, reliability in these explanatory features grows increasingly important for maintaining developer confidence and adoption rates.
8The Importance of AI Transparency
This incident underscores the critical role of transparency and accountability in AI systems. Understanding an AI's decision-making process builds user trust, particularly for developers implementing these tools in production environments. Companies like Anthropic must balance performance with user experience while maintaining open communication with their developer community. Such challenges represent growing pains in the maturing AI industry as it establishes best practices for responsible deployment.
Original article: www.theregister.com - Articles — Claude Code'da Boş Düşünme Blokları Sorunu