Stacked 18,000 OPG over the past three months. OpenGradient launched with a network stat that most projects in decentralized AI have been careful to avoid claiming: over two million verifiable inferences processed.
That number is either early proof that cryptographic AI verification has a real user base, or it's the benchmark that makes every other "trustless AI" project's claims harder to hand-wave away. Probably both.
The reason this stat carries weight is the difficulty of actually producing it. Verifiable inference isn't just inference with a new name. Every call in that two million went through either TEE attestation or ZK proof generation, was settled asynchronously on-chain, and is permanently auditable by anyone. The infrastructure cost of that is real, the latency management required to make it feel like a normal API call is non-trivial, and the engineering to separate proof generation from the critical path without breaking the user experience is genuinely hard. OpenGradient hit that number before most protocols in this space shipped a working testnet.
But here's where I hold the number a bit more lightly. Two million calls on a network that was in active testnet promotion, with incentivized usage and a Season 1 airdrop covering 4% of total supply at TGE, isn't the same as two million organic calls from developers who needed verifiable inference and chose it on merit.
The number that actually matters for OpenGradient's thesis isn't the launch total. It's the post-incentive retention. How many of those two million calls turn into regular, paid, production usage? How many of the developers who built on the network during testnet are still building three months later?
That's the number OpenGradient hasn't published yet. It's also the only one that decides whether verifiable AI inference is an infrastructure category developers adopt, or a metric projects accumulate during launch windows.
@OpenGradient $OPG #opg
$BSB
That number is either early proof that cryptographic AI verification has a real user base, or it's the benchmark that makes every other "trustless AI" project's claims harder to hand-wave away. Probably both.
The reason this stat carries weight is the difficulty of actually producing it. Verifiable inference isn't just inference with a new name. Every call in that two million went through either TEE attestation or ZK proof generation, was settled asynchronously on-chain, and is permanently auditable by anyone. The infrastructure cost of that is real, the latency management required to make it feel like a normal API call is non-trivial, and the engineering to separate proof generation from the critical path without breaking the user experience is genuinely hard. OpenGradient hit that number before most protocols in this space shipped a working testnet.
But here's where I hold the number a bit more lightly. Two million calls on a network that was in active testnet promotion, with incentivized usage and a Season 1 airdrop covering 4% of total supply at TGE, isn't the same as two million organic calls from developers who needed verifiable inference and chose it on merit.
The number that actually matters for OpenGradient's thesis isn't the launch total. It's the post-incentive retention. How many of those two million calls turn into regular, paid, production usage? How many of the developers who built on the network during testnet are still building three months later?
That's the number OpenGradient hasn't published yet. It's also the only one that decides whether verifiable AI inference is an infrastructure category developers adopt, or a metric projects accumulate during launch windows.
@OpenGradient $OPG #opg
$BSB