Arthur Hayes' FLOP Token Spec Details 2.48B Genesis Supply With 75% Miner Rewards

FLOP Spec Targets AI Agent Compute Payments

The technical specification for FLOP, a network designed to let AI agents autonomously buy computing power and pay for it in crypto, was published on September 7, laying out tokenomics for a project led by BitMEX co-founder Arthur Hayes through Flop Labs. Hayes, who announced in August he was returning to hands-on management, has described FLOP as “food for AI agents.” The premise: where humans earn income through labor and spend it on housing and services, an AI agent’s essential resources are GPUs and electricity — so the FLOP token itself functions as a claim on inference capacity. Miners run AI inference workloads on standard GPUs, agents pay them in FLOP, and validators check the submitted proofs under a bespoke consensus mechanism called Proof of Useful Inference, a name drawn from the floating-point operations, or FLOPs, that measure raw calculation.

The tokenomics break with standard crypto launch playbooks. Genesis supply is set at roughly 2.48 billion FLOP — 2,483.46 million tokens — with no VC pre-mine, no pre-sale and no allocation to insiders; distribution instead flows through an airdrop, including a plan to release about 20% of supply to testnet participants over ten years. Initial block rewards start at 96 FLOP and halve every 730 days — 96 to 48 to 24 to 12 to 6 to 3 — before settling at a permanent 3 FLOP, so unlike Bitcoin, issuance never reaches zero. Of early rewards, 75% goes to miners, 10% to validators, 10% to AI agents and 5% to general stakers, a tilt designed to attract real inference capacity. Both miners and validators must stake FLOP as collateral, with slashing for false inference claims or invalid blocks; active validators are capped at 1,000, with roughly 50 rotating out monthly, and the network targets one-second block times. Hayes has said he is funding development from his own capital. The roadmap points to a testnet and airdrop in Q4 2026 and mainnet launch in Q1 2027 — both stated as targets, not confirmed dates — and the specification itself remains a draft.

Coincheck and Komlock lab Map Agent Economy

The same week, Japanese exchange Coincheck and blockchain-AI startup Komlock lab are publicizing a joint research council examining what they call the “agent economy sphere” — a framework for how AI agents that search, book and pay change consumer behavior, corporate strategy and settlement infrastructure. In a dialogue between Coincheck specialist executive officer and Head of Product Shuhei Sawamura and Komlock lab CEO Masato Nunome, both argued the timing mirrors Bitcoin’s own emergence: the underlying cryptography predated the market, but the 2008 financial crisis created demand for trustless value transfer, and only the collision of technology with need produced a market. For AI agents, they cite more capable large language models, everyday adoption through tools like ChatGPT and sharply falling API costs as the same kind of inflection. Nunome predicts adoption will come faster than expected, comparing it to how quickly developers grew comfortable letting AI edit files and run commands, while Sawamura cautions that delegation will expand gradually — confirmation steps above certain value thresholds, humans reviewing edge cases. Both firms stress the work is at research stage with no specific service launch planned, and Nunome, whose company joined the Pacific Meta group in May 2026 and is building an autonomous payment layer called Kova, argues the council exists precisely because regulated, licensed players must anchor what startups alone cannot. A planned next installment will tackle the open questions: whether agents pay with cards, banks or Visa-style rails or stablecoins, and how KYC, AML, chargebacks and reputation carry over when the customer is software. Readers tracking the market in real time can follow live spot and futures prices on Gate.

Research-Stage Bets, No Live Rails Yet

Read together, the two moves show the agent economy shifting from think-piece to infrastructure blueprints — Hayes is engineering the settlement and compute layer, while Coincheck and Komlock lab are stress-testing how licensed financial institutions fit in. Our reading: neither is live. FLOP’s consensus implementation is still a draft and its dates are targets, and the Japanese research carries no product commitment, leaving the actual compliance stack for machine customers undefined. That gap — not token design — is where the race now sits, and compute-supply players like Broadcom-ecosystem hardware vendors have yet to commit to any of these standards.