Summary
On September 9, 2026, Cloudflare AI Gateway extended its custom cost metrics so the cf-aig-custom-cost header now accepts per-cache-read-token and per-cache-write-token rates. Teams can reflect negotiated cache pricing across providers in AI Gateway's spend tracking instead of approximating cache costs with standard input/output rates.
What changed
AI Gateway's cf-aig-custom-cost header gained cache-read and cache-write token rate fields on September 9, 2026, letting operators encode cache-tier pricing for accurate per-request cost accounting.
Why it matters
Prompt caching now materially changes LLM bills, and gateways that cannot price cache reads and writes separately understate or overstate spend. Accurate cache accounting lets platform teams attribute and cap costs correctly as caching-heavy agent workloads scale.
Evidence excerpt
AI Gateway custom cost metrics now support cache-read and cache-write token rates via the cf-aig-custom-cost header, enabling negotiated cache pricing across providers.