Better prompt caching for GPT-6
GPT-6 prompt caching now applies discounts to eligible shared prefixes reused within 30 minutes.
OpenAI says the GPT-6 family ships with higher cache hit rates by default, aimed at long-running agents that carry instructions, tools, and context across many API calls. Developers get a new dashboard for cache hit rates, diagnostics for missed reuse, and explicit breakpoints to choose which prompt prefixes should stay cached. The note also says reasoning effort can be adjusted between responses without breaking cache, if handled through the new configuration update path. OpenAI's note
OpenAI says the GPT-6 family ships with higher cache hit rates by default, aimed at long-running agents that carry instructions, tools, and context across many API calls. Developers get a new dashboard for cache hit rates, diagnostics for missed reuse, and explicit breakpoints to choose which prompt prefixes should stay cached. The note also says reasoning effort can be adjusted between responses without breaking cache, if handled through the new configuration update path. OpenAI's note
score 5