Anthropic's Claude models, particularly newer versions like Claude Sonnet 5.5 and Opus 5.5, incur costs for "thinking tokens" which are billed at the higher output token rate. These thinking tokens are generated by default and their text is often empty unless a summary is explicitly requested, leading to potential unnoticed expenses. Users can manage these costs by adjusting the "effort" setting, which controls the depth of the model's thinking process and can reduce token usage for simpler tasks, or by ensuring `max_tokens` is set appropriately to avoid costly retries. AI
IMPACT Users can optimize AI spending by understanding and managing "thinking token" costs on Anthropic's Claude models.
RANK_REASON The item discusses cost management and configuration of an existing AI model's features, rather than a new release or research.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →