Anthropic rows do not state temperature (sampling) support #496

Problem

The anthropic rows never state temperature in capabilities. All 14 current rows list reasoning, reasoning_effort, structured_outputs, response_format, max_tokens, tool_choice, and tools, and nothing about sampling. An absent key is unknown, so a reader cannot tell which Claude models take temperature, top_p, and top_k.

This matters because the newest models reject non-default sampling with a 400 (Claude Sonnet 5, Opus 4.7 and later, Fable 5). Older ones take it (Sonnet 4.5, Haiku 4.5). The openrouter rows list temperature: true for every Claude model, including the ones that reject it, so that is not a source either.

TanStack AI's native sync (TanStack/ai#1308) reads only the native row now. Without this fact it types every new Claude model as max-tokens-only.

Ask

• State capabilities.temperature (and top_p, top_k if the docs say) for each anthropic row: false where Anthropic's docs say sampling is rejected, true where it is accepted.
• Record the source in factSources.capabilities.

Acceptance

• claude-sonnet-5-5, claude-opus-5-5, and claude-fable-5-1 have capabilities.temperature === false.
• claude-sonnet-4-5-20250929 and claude-haiku-4-5-20251001 have capabilities.temperature === true.