One prompt. The right model.
A receipt for every answer.
Ping reads your prompt, picks the model that fits it best, shows you exactly why, and logs a receipt that’s posted on-chain. Holders get more requests and stronger models, with the rules written in a contract instead of a hidden server setting.
Candidates
quality for this task · live latency · cost · score| Model | Quality | Latency | Cost | Score |
|---|
Trace
Task type
from the promptYour access
read from PingLedgerTiers
| Tier | Hold | Requests/day | Adds |
|---|---|---|---|
| Visitor | 0 | 10 | Pico, Swift, Lingo |
| Holder I | 10K | 30 | Coda, Atlas |
| Holder II | 100K | 80 | Sage XL |
| Holder III | 1M | 200 | Side-by-side compare |
Route, answer, prove it
The router is off-chain and fast. What it promises is on-chain and checkable.
Pick by fit
The prompt is classified (code, reasoning, writing, translation, data, chat), then each model is scored on its arena quality for that task, live latency and cost. Down or degraded models are skipped, with automatic fallback.
Log every answer
Each answer gets a receipt: hashes of the prompt and reply, the model, latency and cost. Every 10 minutes the batch’s Merkle root goes on-chain. The prompt and the answer never do.
Access in a contract
How many requests a wallet gets is read from PingLedger: fixed thresholds, fixed daily credits. After that, pay per request in USDG via x402.
Why not just a router behind a token check?
| Typical token-gated router | Ping | |
|---|---|---|
| Holder rules | Hidden server setting | Thresholds and credits in a public contract |
| Model choice | Fixed formula on keywords | Arena quality per task, live latency, cost, all shown |
| Proof of service | None | Receipt per answer, Merkle root on-chain every 10 min |
| Out of free requests | Wait until tomorrow | Pay per request via x402, no account |
| Provider down | Error | Automatic fallback; failures don’t use credits |