GPT-6 prompt caching cuts repeat-token costs by up to 90%
GPT-6 now reuses eligible shared prompt prefixes for 30 minutes, cutting cached input-token costs by up to 90% and reducing latency for persistent agents that send repeated context. Developers can track hit rates in a new dashboard, then investigate misses with diagnostics that identify changes such as modified tools and estimate affected tokens. OpenAI says GitHub Copilot cut the share of tokens needing fresh processing by more than 50% across billions of requests versus its previous baseline.