Paying for AI compute stopped being the hard problem on July 14, when Visa, Mastercard, and the Linux Foundation put their names on x402.

We're partnering with 0G Labs because the harder problem, proving the compute you paid for actually ran, still needed solving.

On the private tier, models run in Intel TDX and H100 or H200 enclaves. The enclave signs the response and the attestation is fixed at deployment, so you check the model hash instead of trusting anyone. Route to Claude or GPT-5.6 and 0G says plainly that's authenticated delivery, not in-chip execution, because it doesn't hold those weights.

Cluster picks the tier per request. Cheap answers route for speed. Numbers that land in a financial report route to attested compute, and the x402 payment on Base only clears once the attestation checks out.

An agent stops hoping the other side was honest. It only pays once it can check.

Full Breakdown Here.