Scan to join WeChat groupAPI Usage Audit
Use controlled samples to inspect cache usage, hidden input overhead, context integrity, and reasoning-effort behavior.
Never runs inside the free authenticity check
The key is used only for this request and is not written to an audit record
A cache signal is not a billing-fraud verdict without the relay invoice
Anthropic sends up to 6 primary samples plus 1 no-instruction control for applicable models; OpenAI sends up to 4. All share 50 seconds. Each request includes a 24,000-character prefix and up to 8 output tokens; the extra control may incur full input cost.
Audit setup
Configure one model endpoint
Audit evidence
Usage samples will appear here after the repeated requests finish
What this result does not prove
This page verifies whether provider-style usage fields expose cache creation or reads. It cannot prove the relay passed a discount to you until its invoice, balance change, or billing export is compared against these samples.