I've been poking around OpenGradient for a bit now. Their network splits things up in a way that actually makes sense for AI work. Inference nodes handle the heavy model runs on GPUs or TEEs. Full nodes just check the proofs later. No forcing every validator to rerun giant models like a regular blockchain. That keeps it fast while still verifiable.
The incentives feel pretty tied together. You pay with OPG for each inference. Node runners earn when they deliver honest results. Creators get paid every time someone uses their model on the hub. Stakers back the validators who verify everything, with rewards over time and slashing if they cheat. No crazy inflation dumping on holders later. Vesting on most tokens pushes people to think longer term instead of flipping quick.
It lines up usage with value. More real apps and agents calling models means more OPG flowing through payments and rewards. But liquidity is still thin early on. Adoption hinges on devs picking verifiable outputs over quick APIs. If TEEs get gamed or proofs stay too slow for big stuff, friction builds.
Overall, it nudges participants toward keeping the network reliable rather than gaming short term. Not perfect, but better aligned than some AI crypto experiments I've seen.
How do you see the split between fast TEE paths and stricter proofs holding up as more trading bots or agents jump in? Will it draw serious builders or stay mostly experiments?