Field notes, written as I ship. Newest first.
96 posts
disable_parallel_tool_use means at most one tool call under tool_choice auto, and exactly one under any or tool. Claude Opus 5.5, Fable 5.1 and Mythos 5.1 reject any and tool with a 400, so the exactly-one contract is not expressible on them at all.
GET /v1/models returns a capability tree and no lifecycle field at all. Meanwhile two of the five platforms that serve Claude run a different retirement clock, and their model tables still list identifiers the Claude API retired months ago.
A Claude refusal is an HTTP 200 with a fallback subsystem behind it. The handoff is marked by a content block, and after the first fallback the routing goes sticky and that marker disappears while the fallback model keeps answering.
Enabling citations does not give you one citation shape. It gives you four, split across two documentation pages, indexed from two different bases, with the source field under two different names.
The Files API left beta, but the old header did not become inert. It now selects which response shape you get, and the one field it hides is the one that tells you a file has stopped working.
The anthropic-ratelimit headers are not one meter. Input is an estimate that gets corrected, output is a running tally, and the tokens header reports whichever limit currently binds.
Structured outputs and strict tool use turn your JSON schema into a separate cached artifact. It has its own 24 hour lifetime, invalidation rules that invert what you would guess, and a retention boundary that zero data retention does not cover.
Context engineering on the Claude platform is not a setting. It is four mechanisms at four points in the pipeline, and the one everybody enables first pays for context window with prompt cache.
A task budget is not a per-task cost ceiling. It resets whenever you send a user message carrying no tool results, which is exactly what most harnesses do when Claude finishes gracefully.
Context rot is real and Anthropic now names it in its own docs. The useful question in 2026 is narrower: your Claude model ships with one of two self regulation mechanisms, or with neither, and they do not overlap. Sonnet and Haiku get context awareness free and automatic. Opus needs a beta header. Opus 4.6 gets nothing.
Comparisons of Claude Opus, Sonnet and Haiku score capability, speed and price per token. None of them mention that the model string is part of your request contract. Swap it and thinking can turn itself on, your answer can move out of content[0], and a max_tokens you never touched can start truncating. Eight measured differences, September 2026.
