Model capability, evaluation integrity, and what inference actually costs.
Prompts between 512 and 1,023 tokens — written off as uncacheable for two model generations — now cache on Claude Opus 5, Fable 5 and Mythos 5. At a tenth of the input price per read, the smallest documentation change in the Claude 5 launch may be the one worth re-benchmarking first.
Anthropic's most capable generally available model is the first to refuse zero-data-retention customers outright: organisations with a ZDR arrangement get a 400 on every Fable 5 request unless they carve out a workspace that retains data for 30 days. For regulated buyers, the model choice is now a data-policy choice.
Claude 5 models think by default; the reasoning bills as output and counts against max_tokens. What flips when you swap the model string.
