Claude Code returns blank thinking blocks, but reasoning still costs you

Anthropic's Claude Code API has a display bug: users requesting reasoning summaries for Opus 4.8 and Sonnet 5 models are getting blank thinking blocks, per The Register, while still being charged for the underlying tokens. Anthropic's documentation clarifies that developers pay for thinking tokens regardless of whether summaries are returned—but some developers report truncated summaries and stream interruptions during long-running sessions. The company says the issue is not widespread and is under investigation.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton