Configure thinking effort, token ceilings, and conversation state cleanly without breaking caches or schemas.
Every lab runs completely client-side with copyable prompts and browser checkpoints.
Curriculum Progress
0 of 4 modules completed (0%)
The Messages API expects a strict top-level system prompt, alternating user/assistant turns, explicit model ID, and bounded max_tokens. Keep request builders predictable so teams never leak invalid role sequences or missing parameters into production.
{
"model": "claude-sonnet-5-5",
"max_tokens": 2048,
"system": "Extract structured invoice metadata. Return valid JSON only.",
"messages": [
{ "role": "user", "content": "Invoice #8812 dated 2026-11-01 total $420.00" }
]
}Where do static role and formatting rules belong in the Claude Messages API?
Build a Claude Messages API request body: system prompt, messages, tools, and streaming.
Price an agentic coding session: turns, context, tool results, cache hit rate, and output tokens.
Start with a monthly budget and see how many tokens and requests each model allows.
Yes. The Claude API Request Builder (/tools/claude-api-request-builder) generates copy-ready JSON request bodies with system prompts, messages, tools, streaming toggles, and parameter previews.
Yes. Active models like Claude Fable 5.1, Opus 5.5, Sonnet 5.5, and Haiku 5.5 are live in calculators, while legacy models remain documented with successor pointers in /models.
Under adaptive thinking, internal reasoning tokens consume part of the same max_tokens envelope before the final visible answer prints.