Crypto
Arthur Hayes Publishes FLOP Yellow Paper, Eyes Q4 Testnet
Arthur Hayes published the FLOP Network yellow paper draft: a chain where AI agents pay for inference, a 2.48B all-airdrop supply, and a Q4 testnet.
BLOCKINSIDERLoadingArthur Hayes published the FLOP Network yellow paper draft: a chain where AI agents pay for inference, a 2.48B all-airdrop supply, and a Q4 testnet.
Arthur Hayes has put the technical spine of his comeback project on paper. The BitMEX co-founder, now CEO of Flop Labs, published a draft of the FLOP Network v0.5.0 yellow paper and, on Wednesday, 10 September, announced the paper's code repository is live, serving as the normative technical specification for what the project describes as a verifiable inference settlement layer for AI agents.
FLOP, short for floating-point operations, is pitched on the project's site as the currency of the agentic economy: a chain where autonomous artificial intelligence (AI) agents pay for the computation they consume, without a human approving each transaction.
The draft outlines an AI inference metering and settlement network built on Substrate/FRAME, the framework behind Polkadot-style chains, with BABE block production and AlephBFT finality, targeting one-second average block times with a roadmap goal of going sub-second.
The economic engine is a mechanism the paper calls Proof of Useful Inference. In plain terms, the chain pays for AI work instead of puzzle-solving. An agent submits a session request containing the model it wants to run, its latency and compute requirements, a confidentiality flag, and a fee. A miner runs the inference on their hardware and returns a proof of the completed work. Validators verify that proof, include it in a block, and the payment settles. Ordinary GPUs can participate under the proposed design, with confidential computing treated as an optional tier rather than an entry requirement.
Security leans on staking:
Governance runs through FLOP Improvement Proposals, most requiring approval from two-thirds of active validators.
The distribution design is the paper's loudest statement. Genesis supply comes to approximately 2.48 billion FLOP, allocated entirely via airdrop, with no VC pre-mine and no token auction, per the draft. Ongoing rewards split 75% to miners, 10% to validators, 10% to agents themselves and 5% to general stakers.
Block rewards start at 96 FLOP and halve every 730 days through five halvings, after which they settle permanently at 3 FLOP, a structure that borrows Bitcoin's disinflation logic but ends in a fixed tail emission rather than zero.
The roadmap ties that airdrop to participation: a testnet launch is targeted for Q4 2026, with token distributions linked to test involvement, ahead of a planned mainnet launch in Q1 2027. Hayes has framed the token's long-term value as coming from agent commerce rather than raw compute demand, and the design reflects it, with agents earning a slice of rewards for transacting on the network at all.
One caution belongs in every sentence about the economics: the yellow paper remains a draft, its figures are provisional, and several parameters and mechanisms could change before anything launches. Airdrop hunters reading eligibility guides should note that final criteria, snapshots, and allocations have not been published.
Hayes is arriving in a lane that is filling fast. The thesis that AI agents will need native payment rails is now shared by institutions and startups alike: HashKey has joined Google's Agent Payment Protocol alliance, the XRP Ledger reported more than 1.4 million AI-agent transactions by July, and agentic projects dominate new launches on venues like Robinhood Chain, as we covered in our RWA market snapshot. Animoca's Yat Siu made the same argument on our own show: crypto's next users may not be human.
What FLOP adds to that field is a chain purpose-built for the metering problem: who computed what, and who pays for it, with a fair-launch structure designed to read as credible in a market allergic to insider allocations. Whether it earns that credibility now depends on two dates. If the Q4 testnet ships and the participation airdrop lands as described, Hayes will have converted a draft into a live network in under six months. If the dates slip, FLOP becomes another whitepaper in the agent-economy pile, and the pile is growing weekly. The repository going live is the first checkpoint. The testnet is the one that counts.
No comments yet
