Last year, I connected an AI risk control module for a friend's company, and after running the model for two weeks, it suddenly became inaccurate. I checked the logs thoroughly and found nothing unusual. The customer service just shrugged: "The model is internal, you can't see it." At that moment, I realized – it’s not about what the AI outputs, but why you trust it.
OpenGradient's mission is to turn "Do you trust me?" into "Can you verify me?".
Their HACA architecture separates execution and verification – the inference nodes run the model off-chain, producing results in milliseconds, while the full nodes are only responsible for validating cryptographic proofs without rerunning the model. Verification is tiered: TEE relies on Intel's SGX hardware for endorsement, which is sufficient for daily use and is the default option for LLM inference; ZKML employs mathematical proofs, offering a high-security ceiling, but generating the proofs might take longer than running the model itself. The white paper doesn’t claim "absolute security" but provides a "trust menu" – you pick efficiency or safety, your choice.
The team's background is indeed solid. CEO Matthew Wang was previously a research engineer at Two Sigma, and CTO Adam Balogh was the former tech lead of Palantir's AI platform. a16z crypto led a $9.5 million investment, with Coinbase Ventures and SV Angel also on board. The mainnet launched on the Base chain on April 21, currently hosting over 4,400 models, processing over 2 million inferences, and validating over 500,000 proofs. Upbit also followed suit with a launch on June 15.
But the numbers need to add up.
TEE relies on the credibility of Intel hardware, and SGX has been attacked via side-channel exploits multiple times. Relying on a single chipmaker's closed-source firmware for the security foundation of verifiable AI is itself a compromise. ZKML is absolutely secure but slow – the project team knows this too, and enforcing ZKML in large-scale scenarios would lead to bottlenecks.
Now, looking at the token. Total supply is 1 billion, with only 190 million in circulation, less than 20%. Ecosystem tokens will be released linearly over 60 months, with staking rewards of 10% spread over 96 months. On June 21, an additional 9.13 million tokens from the foundation will be unlocked. This isn't a dumping signal, but the supply is indeed increasing. The top 10 wallets hold 94.2% of the circulating supply, indicating extreme concentration of chips. The price has retraced from its historical high of 0.48 to around 0.16. @OpenGradient
I believe in the direction of verifiable AI. But the real challenge in this space isn’t whether the technology can work, but whether anyone is willing to pay a premium for the three words "verifiable AI"!
#OPG $OPG
OpenGradient's mission is to turn "Do you trust me?" into "Can you verify me?".
Their HACA architecture separates execution and verification – the inference nodes run the model off-chain, producing results in milliseconds, while the full nodes are only responsible for validating cryptographic proofs without rerunning the model. Verification is tiered: TEE relies on Intel's SGX hardware for endorsement, which is sufficient for daily use and is the default option for LLM inference; ZKML employs mathematical proofs, offering a high-security ceiling, but generating the proofs might take longer than running the model itself. The white paper doesn’t claim "absolute security" but provides a "trust menu" – you pick efficiency or safety, your choice.
The team's background is indeed solid. CEO Matthew Wang was previously a research engineer at Two Sigma, and CTO Adam Balogh was the former tech lead of Palantir's AI platform. a16z crypto led a $9.5 million investment, with Coinbase Ventures and SV Angel also on board. The mainnet launched on the Base chain on April 21, currently hosting over 4,400 models, processing over 2 million inferences, and validating over 500,000 proofs. Upbit also followed suit with a launch on June 15.
But the numbers need to add up.
TEE relies on the credibility of Intel hardware, and SGX has been attacked via side-channel exploits multiple times. Relying on a single chipmaker's closed-source firmware for the security foundation of verifiable AI is itself a compromise. ZKML is absolutely secure but slow – the project team knows this too, and enforcing ZKML in large-scale scenarios would lead to bottlenecks.
Now, looking at the token. Total supply is 1 billion, with only 190 million in circulation, less than 20%. Ecosystem tokens will be released linearly over 60 months, with staking rewards of 10% spread over 96 months. On June 21, an additional 9.13 million tokens from the foundation will be unlocked. This isn't a dumping signal, but the supply is indeed increasing. The top 10 wallets hold 94.2% of the circulating supply, indicating extreme concentration of chips. The price has retraced from its historical high of 0.48 to around 0.16. @OpenGradient
I believe in the direction of verifiable AI. But the real challenge in this space isn’t whether the technology can work, but whether anyone is willing to pay a premium for the three words "verifiable AI"!
#OPG $OPG