Newsroom
8 September, 2026 / News / AI / Tags: flop, inference, miners, validators, genesis

BitMEX co-founder releases technical specifications for a proof-of-useful-inference chain with 2.48 billion genesis tokens distributed solely via airdrops
Arthur Hayes, co-founder of BitMEX, has published the technical paper for the FLOP Network, outlining a blockchain designed for autonomous AI agents to purchase computing resources directly from miners. The protocol introduces a consensus mechanism called proof of useful inference, or PoUI, under which miners earn rewards by executing AI model inferences rather than performing traditional cryptographic hashing.
FLOP, short for floating-point operations, functions as the network’s native currency. Agents submit session requests to the mempool specifying model-weight hashes, latency limits, computational requirements measured in FLOPs, confidentiality preferences and the fee offered. Miners with suitable hardware accept these requests, complete the inference over a private connection and return a proof of the work. Validators then incorporate the proof hash into blocks to settle the transaction, after which miners receive both the agent’s session fee and a share of network block rewards.
The specifications set the genesis supply at approximately 2.48346 billion FLOP tokens. The entire initial allocation is designated for airdrop distribution, with no venture capital premine, presale or token auction. This approach continues the fair-launch stance Hayes described earlier when announcing his return to an operating role at Flop Labs.
The project previously indicated a large airdrop targeted for the fourth quarter of 2026, followed by a genesis block in the first quarter of 2027. The newly released documentation provides the precise token figures and emission schedule that earlier announcements had left open.
Block rewards begin at 96 FLOP and follow a fixed halving schedule every 730 days. After five successive reductions the subsidy stabilizes permanently at 3 FLOP per block. The sequence runs 96, then 48, 24, 12, 6 and finally 3 FLOP.
Rewards are divided among four participant groups: miners receive 75 percent, validators receive 10 percent, agents receive 10 percent and ordinary stakers receive the remaining 5 percent. Miners also collect the inference fees paid by agents, creating two distinct revenue streams tied directly to useful computational work.
The network targets an average block time of one second with deterministic sub-second finality. Development plans aim for block production below one second.
Both miners and validators must stake FLOP tokens as collateral. Dishonest behavior triggers slashing: miners face penalties for misrepresenting completed inference, while validators risk sanctions for publishing invalid blocks. Severe violations can result in full forfeiture of staked tokens and expulsion from the network.
Token holders who do not operate infrastructure may delegate their stake to miners or validators and receive a proportional share of rewards. The active validator set is limited to 1,000 participants. Approximately 50 validators are expected to rotate each month according to verified workload and uptime. Stake weight is planned to influence ranking once a participation threshold is met.
Validators also hold governance authority. Protocol changes advance through FLOP Improvement Proposals, most of which require approval from two-thirds of the active validator set before activation.
Model weights reside in a data-availability layer while validators assemble blocks containing hashes of the inference proofs submitted by miners. Ordinary graphics processing units can join the network; confidential computing is treated as an optional higher tier rather than a mandatory requirement.
The design positions FLOP as payment for verifiable AI inference, enabling autonomous agents to acquire compute resources on demand. The technical paper remains labeled a draft and is subject to further refinement ahead of the planned network launch.









