OpenAI's API dashboard now tracks your prompt-cache hit rate
Cached input is 90% cheaper but writes cost 1.25x, so read/write ratio is real money — now visible. Plus: Vercel Functions can emit custom metrics.
Copy markdown
See whether your caching actually pays off
OpenAI shipped a Prompt Caching dashboard under Platform → Usage that charts cache-hit rate over time, reads per write, and the split of cache-read, cache-write and uncached tokens, filterable by model and service tier. Cached input bills at 0.1x the base rate (90% off) but writes on GPT-5.6+ cost 1.25x, so a low reads-per-write ratio means you're eating the write penalty without banking the discount.
Vercel Functions can now emit custom metrics
Vercel added custom metrics to Observability, so a Function can push its own counters and gauges and chart them next to Vercel's built-in data. For AI endpoints that's a clean way to track per-request token spend, model latency, or cache hits without wiring up a separate telemetry stack.