**Anthropic prompt caching appears inactive through Vercel AI Gateway** #4784

Anthropic prompt caching appears inactive through Vercel AI Gateway

My custom provider uses Vercel AI Gateway’s Responses API. Across four sessions, 138 anthropic/claude-opus-5.5 requests reported zero cache-read and cache-write tokens, including requests seconds apart with approximately 60,000-token inputs.

These counts come from Desktop’s usage events; outgoing request bodies and Gateway cache usage haven’t been independently verified.

Please investigate whether Desktop sends explicit cache markers or providerOptions.gateway.caching: "auto", and whether cache usage is reported correctly. I couldn’t find a supported provider setting to enable this.

Documentation: https://vercel.com/docs/ai-gateway/models-and-providers/automatic-caching