No.284
OpenGradient OPG
the lie detector oracle
🎭 a watchful familiar that runs whatever brain you ask, then stamps the answer so it can't pretend it used a different one
💬 “Ask me anything and I'll run the model you picked, no swaps behind your back. When I hand back the answer, a proof rides along with it. Check the proof, and you'll see exactly which brain I used and that not a single word was touched.”
- The job: a decentralized AI network that runs models for apps and on-chain agents, and proves the result is honest.
- The trick: heavy math runs off-chain on GPU and TEE nodes; a zkML proof or TEE attestation settles on-chain so anyone can verify it.
- The token: OPG pays for inference, secures the network through staking, and votes on governance.
- Supply: fixed at 1,000,000,000 OPG — capped, not minted forever. Launched April 21, 2026.
📖 The Story
Most AI today is a black box. You send a prompt, an answer comes back, and you simply trust that the company ran the model it promised and didn't quietly edit the output. For a chatbot that's fine. For a trading agent moving real money, or a contract that pays out based on an AI's verdict, blind trust is a problem.
OpenGradient calls itself the network for open intelligence, and its whole purpose is to remove that blind trust. It hosts AI models and runs them for whoever asks, but it does one extra thing the big AI companies don't: it produces a cryptographic proof for every answer. The proof lets anyone confirm which model actually ran, on what prompt, and that the output came back untouched.
On April 21, 2026 the project launched its OPG token through a Binance Wallet and PancakeSwap event. By then it had raised about $9.5 million from backers including a16z Crypto and Coinbase Ventures, with advisors like Balaji Srinivasan, NEAR's Illia Polosukhin, and Polygon's Sandeep Nailwal lending their names. News reports name Matthew Wang (formerly of the quant fund Two Sigma) and Adam Balogh (formerly head of AI at Palantir) as its co-founders.
Its whole pitch sits on one stubborn promise: an AI that can be checked, not just believed.
📊 Stats
These bands are altrookie's own beginner-friendly read, not market data.
🧩 How it works
The clever part is splitting the work in two. Running a big AI model is far too heavy and slow to do directly on a blockchain, so OpenGradient does the math off-chain on a network of GPU and oracle-style nodes, including TEE nodes (a TEE is a sealed, tamper-resistant slice of a chip). Only the small proof goes on-chain, where it's cheap to verify. You get web-speed answers and on-chain honesty at the same time.
The OPG token is the fuel for all of this. You spend it to pay for inference, node operators stake it as a security deposit and earn rewards for honest work, and holders use it to vote on how the network is run.
🌗 Light & Shadow
- Solves a real, growing pain: as AI agents start handling money on-chain, provable answers matter more than fast ones
- Supply is hard-capped at 1 billion OPG, so the token isn't quietly diluted over time (about 19% was circulating as of June 2026)
- Serious backers and advisors (a16z Crypto, Coinbase Ventures; advisors from NEAR and Polygon), which buys it attention and runway
- Very young. The token only launched in April 2026, so its track record under real load is thin
- zkML proofs are expensive to compute and still early tech; proving heavy models cheaply at scale is an unsolved race across the whole field
- Crowded category. It competes with other verifiable-AI projects, and most users may never check a proof (if no one verifies, the headline feature goes unused)
- Founder names come from crypto news, not the official site, and roughly 81% of supply is still locked — both worth watching
🧬 Evolution lineage
OpenGradient isn't a fork of anything and has no hard-fork sibling. It's a standalone token built on Base (an EVM chain), so its kinship is by theme, not by code: it sits in the decentralized / verifiable AI cohort.
🧭 Meet other friends
❓ FAQ
- What is OpenGradient?
- A decentralized AI network. It hosts and runs AI models, and crucially it attaches a cryptographic proof to each answer, so you can check which model produced a result and that nobody altered it afterward. It is built to fix the AI ‘black box’ trust problem.
- How does OpenGradient prove an AI answer is honest?
- The heavy model math runs off-chain on GPU and TEE nodes for speed. The network then produces a proof — a zkML proof or a TEE attestation — and settles it on-chain. Anyone can check the proof, so the result is trustworthy without re-running the whole model yourself.
- What is the OPG token used for?
- Three things: you pay OPG to run AI inference on the network, node operators stake OPG and earn rewards for keeping it running, and holders use OPG to vote on governance. Supply is fixed at 1 billion OPG, so it isn't minted endlessly.
- Is OPG its own blockchain?
- No. OPG is an EVM token, not an L1 and not proof-of-work. Base is its reference chain, and it is also deployed on Mantle and BNB Smart Chain, bridged across them via LayerZero. The AI compute happens on a separate node network, not by mining.
- When did OPG launch and who is behind it?
- OPG launched on April 21, 2026 through a Binance Wallet and PancakeSwap event. The project raised about $9.5M from backers including a16z Crypto and Coinbase Ventures. It is reported to be co-founded by Matthew Wang (ex-Two Sigma) and Adam Balogh (ex-Palantir).
⚠️ Not investment advice. All figures are for information only