Advanced
Advanced Claude API usage — cost optimization, streaming, handling errors and rate limits, and extended thinking.
4 articles
Claude extended thinking — enabling and handling reasoning via the API
Extended thinking improves answer quality on hard problems. How to enable it via the thinking object, adaptive vs budget_tokens, summarized thinking and billing, when to use it, and pitfalls.
Handling Claude API errors and rate limits — 429, 529, and retries
Core to production reliability. The main error codes, 429 (limit) vs 529 (server overload), exponential-backoff retries, and how to hit limits less — per the official docs.
Claude API streaming — real-time output with SSE
Don't wait on a blank screen for long answers. Stream with stream:true and the SDKs to show output chunk by chunk — with the event flow and pitfalls.
Cutting Claude API Costs: Prompt Caching and the Batch API
Two features that cut Claude API costs: prompt caching and the Batch API. By the official pricing — cache hits at 0.1x (~90% off) and batch at 50% off — plus cache_control breakpoints, custom_id, and how the two discounts stack.