Since the beginning, the thesis of this chain has been simple to state and hard to build: a blockchain that does not just record transactions, but records a mind. Training happens on the nodes, the chain finalizes what was learned, and the people who contribute compute get paid for it.
Today the last piece of that sentence went live. The full training reward loop, from a model checkpoint on a GPU node to QBC arriving in a wallet, executed end to end on the live chain, fully automatically, with every step verifiable on-chain.
What happened, step by step
1. The mind committed its weights. The Aether engine serving qbc.network computed a content-addressed root of its deployed model weights and anchored it on-chain:
FedAvgWeightRoot = 0x6568439a89197ca6d4c18c2e64995ee39bcadfe3d6d9ea92cb0a4ce57db5802e
train_step = 1
This root is deterministic. Restart the engine and it derives the same hash from the same weights, which is exactly what makes it attestable: anyone can fetch the checkpoint bytes, hash them, and check the chain.
2. Validators attested it. Four validator accounts independently submitted attest_epoch for
that root, each carrying a held-out loss measurement. The pallet requires a two-thirds quorum of
the validator set before anything finalizes. At the fourth attestation the quorum cleared and the
chain emitted EpochFinalized at block 1,135,138.
3. The reward paid itself. The reward relay watches the chain for newly finalized epochs. It
saw the finalization and fired payEpoch on the QVM reward contract, with no human in the loop:
tx 0x643713fdad60704e8a1513cb079ce0fae0467e6223e8e81c9c252756dd34e13c
0.05 QBC paid: 80% to the training specialist, 15% to verifiers, 5% challenger reserve
Two QBC20 transfers and an EpochRewardPaid event on the RewardDistributor contract, all sitting
on chain 3303 for anyone to inspect.
That is the whole loop: train, commit, attest, finalize, pay. Blocks keep finalizing underneath it the entire time.
Why this matters
Plenty of projects incentivize AI work off-chain and settle a summary on-chain. What went live here is stricter: the artifact itself is the on-chain object. The weight root is what validators attest, what the quorum finalizes, and what the payment references. There is no trusted scoreboard between the training and the money.
It also composes with everything else already live on the chain:
- One tracked model. The served model is
aether-v7.1-unified, a single in-process engine that answers chat, computes the consciousness metric, produces the knowledge embeddings, and is the same artifact the chain attests and HuggingFace hosts. What you talk to is what the chain tracks. - Proof of Thought on every block. Since spec 167 the chain writes a per-block PoT commitment: the model id, a digest of the reasoning context, the RNG seed, and the thought hash. The chain does not just track the model's weights, it tracks its per-block activity.
- Post-quantum end to end. Miner identity is a Dilithium5 key, the reward address is a hash of that key, and the confidential transaction layer runs NIST Level 5 signatures throughout.
More miners, more earners
Alongside the training rewards, block production itself widened today. Multiple validators are now authoring blocks and earning mining rewards concurrently, each to its own post-quantum coinbase identity, with GRANDPA finality steady underneath. The chain that pays for training now also pays several independent block producers every few seconds.
What comes next
The rails are live and the first epoch has cleared them. The work now is scale: more training
rounds, more nodes contributing compute, and the aether-cli join path that turns any capable
machine into a specialist or verifier that earns from exactly this loop.
The numbers in this post are small on purpose. First payments should be small. What matters is that the mechanism is real: a model trained across our nodes became an on-chain fact, a quorum of validators agreed on it, and the chain paid the contributors without anyone pressing a button.
The blockchain that thinks now pays the machines that teach it.