OpenAI launches a Prompt Caching dashboard on the API platform

OpenAI has released a Prompt Caching dashboard on the API platform, giving developers a direct view into cache hit rates, cache reads per write, and the token breakdown between cache-read, cache-write, and uncached usage — visibility that was previously only inferable from raw usage logs.

What's new

Per the OpenAI API changelog entry dated August 20, 2026, the new dashboard lets customers "track your cache hit rate over time, cache reads per write, and the breakdown of cache-read, cache-write, and uncached tokens to understand your caching efficiency and identify opportunities to improve." Metrics can be filtered by model and by service tier, letting teams isolate caching performance for a specific deployment rather than looking at account-wide aggregates. The dashboard lives at platform.openai.com's usage section, alongside OpenAI's other billing and usage reporting tools.

Prompt caching itself — which reuses previously processed prompt prefixes to cut latency and cost on repeated or templated requests — has been part of the API for some time, priced at a discount versus fresh input tokens. Until now, though, confirming whether an integration was actually benefiting from caching meant parsing raw per-request token counts rather than a purpose-built view.

Context

This ships the same week OpenAI cut GPT-5.6 Sol's per-token pricing and added per-request regional processing selection — a cluster of platform refinements aimed at making the API cheaper and easier to operate at scale rather than introducing new model capability. Caching efficiency directly affects the bottom line for high-volume API customers: a low cache-hit rate on a workload architected to rely on caching (say, a chat application resending long system prompts) can mean paying full input-token price on requests that were supposed to be discounted.

Competitors have shipped similar transparency tooling — cloud providers running inference at scale generally expose cache-related metrics for cost governance — and this brings OpenAI's own dashboard tooling in line with what large customers already expect to audit their spend.

Why it matters

For teams running prompt-caching-dependent architectures, a hidden cache-miss problem can be an expensive, silent failure mode: latency and cost both degrade, but nothing in the API response signals that caching stopped working. A dedicated dashboard turns that into something a team can monitor and alert on, rather than discovering weeks later during a billing review.

It's also a small but telling signal about where OpenAI is investing platform effort: not just new models, but the operational tooling — usage visibility, regional controls, pricing — that determines whether existing customers can run larger, cheaper, more reliable workloads on the API they already have.