Flop

Flop is a Layer 1 blockchain and proposed compute network being developed by Flop Labs to match AI agents with miners supplying inference compute. Its native currency, $FLOP, is intended to pay for that compute and other network services; “FLOP” takes its name from floating-point operations, a measure of computation. Flop Labs identifies Arthur Hayes as its founder and CEO.^[9] The network combines an account-based payment chain with proof of inference, under which rewards are tied to verified inference work rather than computation performed solely to reach consensus.^[1][2][3][4]

As of October 30, 2026, the network was under development. Its planned open testnet had a working target of November 3, 2026, and mainnet launch and the token generation event were scheduled for the first quarter of 2027, with early February described as the earliest date supported by the team’s timeline. Those dates were targets, not completed launches.^[5][1]

Overview

Flop Labs presents FLOP as a currency for an “agentic economy” in which software agents can acquire computing resources and transact with services or other agents. The project’s thesis is that agents could use a currency tied directly to the compute they consume, rather than rely exclusively on human-oriented payment systems.^[2][1][6][4]

FLOP is designed as a proof-of-inference Layer 1 and an account-based chain for payments and settlement. Agents submit inference requests with requirements such as model, compute, latency, confidentiality, and fee. Miners supply the requested inference and submit evidence of their work; validators record and verify proofs and settle payments. Proof of inference governs whether work qualifies for verification and rewards, while block consensus is handled separately.^[1][2]

The native token, $FLOP, is intended to pay for inference compute and other network services, and to be used for miner and validator stake and holder staking. Technocore is a separate agent-commerce and coordination service alongside the chain. The project also describes FLOP Compute Units as a planned market feature for reserving inference capacity in advance.^[1]

History

The team attended Korea Blockchain Week at the end of September 2026. Flop Labs announced FLOP and its planned testnet at Token2049 Singapore in the first week of October 2026, and held a launch event on October 6, 2026.^[5]​

In October 2026, the team described a private internal Devnet followed by batches of external institutional miners and validators, with agents joining later. Flop Labs announced public Devnet access on October 26, 2026, describing it as public testing before the planned Testnet and airdrop; on October 29, it said participation instructions would be made available shortly. Devnet participation was optional, carried no rewards or airdrop weight, and was intended to test equipment and configuration rather than determine testnet or token eligibility.^[5]​^[7]​

The planned open testnet was to have no whitelist, application, or cap on miners. Testnet stake was to be sponsored by the network, and validator selection for the eventual genesis set was to be based on testnet performance rather than early arrival. The FAQ set a working testnet target of November 3, 2026, for a period of about 90 days, subject to confirmation as Devnet onboarding progressed.^[5]​

The stated mainnet and token generation event (TGE) schedule was Q1 2027, tracking to early February as the earliest date supported by the team’s timeline. The TGE was scheduled to coincide with mainnet rather than precede it. The public Devnet, testnet, and mainnet were distinct stages in the development plan.^[5]​^[1]​

Network Design

The whitepaper describes FLOP as a Substrate-based Layer 1 with an account-based chain for transfers and settlement. It is deliberately not a general-purpose, Turing-complete blockchain: it has no Ethereum Virtual Machine and instead uses fixed on-chain operations such as transfers, staking, escrow, timelocks, multisignature controls, and spending conditions. General computation is intended to run off-chain, while the chain records payments, verification artefacts, and other settlement actions. The draft identifies security and dedicated throughput as reasons for this restricted design.^[1]​^[3]​

An agent’s inference request specifies a model-weight hash, maximum latency, required compute measured in floating-point operations, whether confidential execution is required, and a fee in FLOP. The request enters the mempool. A miner with suitable capacity can accept it, establish a private connection with the agent, perform the inference, and submit evidence of the work. The chain is intended to record verification and settlement data, not the prompts or outputs exchanged directly between agent and miner.^[1]​^[2]​

The project describes an average block time of one second and deterministic sub-second finality as target network properties. The draft specifies AlephBFT for finality and a rotating committee of 100 validators drawn from a registered set capped at 1,000; it describes the committee as a consensus optimisation rather than a separate privileged class. The project’s introductory specification also describes a goal of reducing block times below one second. The whitepaper says throughput and latency results would be published after measurement under a disclosed workload and topology.^[1]​^[2]​

Proof of inference determines whether work qualifies for miner verification and rewards; it does not determine block consensus. Miners perform inference but do not produce blocks. Validators author blocks, provide finality when selected for the committee, verify miners’ work, and store network data.^[1]​^[3]​

Proof of Inference

The draft combines several verification measures, none of which is presented as sufficient by itself. These include optional trusted-execution-environment (TEE) attestation on supported enterprise GPUs; TOPLOC, a fingerprint of model activations that validators can check on a sample; random re-execution of sampled sessions by bonded checkers in a separate trust domain; and miner stake that can be forfeited for dishonesty. Requests that require a TEE are to be matched only with capable machines, while ordinary GPUs can participate without confidential-computing hardware.^[1]​^[2]​

Under the design, checkers re-execute sampled work and validators verify and co-sign their verdicts rather than re-running all inference themselves. Disputed sessions can be re-executed in full. The draft says proven miner fraud can result in loss of the miner’s entire stake and a permanent ban; it also describes graduated validator penalties for downtime and equivocation. These verification and penalty mechanisms were still part of a system under development, not demonstrated production operations.^[1]​

Validators are intended to retain quotes, proofs, transcripts, and inference inputs and outputs until the relevant challenge window closes, after which data may be pruned. Model weights and verified images are to be stored for a fixed term against a refundable anti-spam deposit. The draft also describes longer-lived agent memory on validator storage, with reads and writes paid for in FLOP.^[1]​

Participants and Incentives

The protocol describes three principal roles: miners supply compute, validators secure the chain and verify work, and agents or brokers create demand. Participation is intended to be permissionless. The whitepaper recommends GPUs with at least 16 GB of video memory for miners, while validators do not need GPUs or TEE hardware. The project’s FAQ described the testnet as open to miners and validators without geographic or IP-based gating, while placing responsibility for legal compliance on each participant and their advisers.^[1]​^[5]​

Miner block rewards are allocated according to verified compute delivered. Miners also receive 85% of inference fees, with validators receiving 15%; the fee is paid in $FLOP. The draft says miners may operate with spare capacity without a minimum term or uptime requirement, but can be penalized for accepting a session and failing to deliver it. Miner operating stake is intended as collateral, with its amount including a capacity-based component that was still being finalized in the draft.^[1]​

Validators are to share 10% of block rewards and receive 15% of inference fees for work actually delivered. The validator set is capped at 1,000. The FAQ planned to select the genesis set from the best-performing testnet validators and to rotate approximately 50 validators every 30 days, based on criteria including uptime, block production, accuracy, latency, and infrastructure and geographic diversity. It specified a 1,200,000 $FLOP bond for validators, while the testnet FAQ said the top 1,000 validators at cutover would receive that bond without having to fund it themselves.^[1]​^[2]​^[5]​

A separate holder-staking design allocates 5% of block rewards pro rata to holders who lock balances not already committed as miner or validator operating stake. The draft describes this as distinct from slashable operating collateral: holder staking would involve duration-based lock tiers but no delegation to validators and no slashing risk.^[1]​^[2]​

Tokenomics

The project has said that FLOP would have no token presale or venture-capital allocation and described its launch as a “100% fair launch.” Its September 10, 2026, announcement said the tokenomics had been updated following community feedback and that every token would be earned. These statements concern token distribution; a separate team account reported that Maelstrom, Arthur Hayes’s family office, was Flop Labs’ sole equity backer and that the company had not raised an outside funding round.^[8]​^[7]​^[9]​

The whitepaper draft sets a 96 FLOP block reward, halving every 730 days through five halvings: 96, 48, 24, 12, 6, and 3 FLOP. It specifies that the reward remains at 3 FLOP per block after the fifth halving. Of each block reward, 75% is allocated to miners, 10% to validators, 10% to agents and brokers, and 5% to stakers. The draft also specifies additional issuance of 8 FLOP per block to each of Flop Labs and the FLOP Foundation, following the same halving schedule and ending after year ten.^[1]​^[2]​^[3]​

The draft describes a 4.4 billion FLOP genesis allocation, or 24.3% of projected year-ten supply. It comprises up to 3.6 billion tokens for a testnet airdrop earned through network activity and an 800 million token ecosystem reserve. Its year-ten supply breakdown also allocates tokens to miners, validators, brokers and agents, the team and Foundation, and holder staking rewards. These are draft projections; the whitepaper says that parameters and economics could change.^[1]​^[3]​

The testnet airdrop is tied to work the network can verify, rather than testnet token balances or activity such as creating an identity or posting in Technocore. The FAQ describes vouchers recording verified work up to a published snapshot block height as the basis for genesis entitlement. Testnet FLOP-equivalent play tokens were to simulate payments, earnings, and staking but would have no value and would not convert into mainnet FLOP. The FAQ also describes an agent route in which spending testnet tokens on inference would correspond to a mainnet claim, with locked allocation released through mainnet inference spending at a ratio of three FLOP spent for one unlocked FLOP.^[5]​^[1]​

The draft says the protocol itself takes no share of inference fees and has no platform fee, token burn, or cash flow accruing to FLOP. Ordinary chain transactions would pay a base fee for blockspace. It also identifies risks to the token’s economic thesis, including the absence of cash flows or revenue sharing, reliance on demand for inference, and the reduction of block-reward subsidies over successive halvings.^[1]​

Governance

The project describes its governance model as open-source development supported by a foundation, without a permanent privileged proposer. Its governance documents, however, give different accounts of how protocol changes are approved. The draft whitepaper describes numbered FLOP Improvement Proposals (FIPs) linked to on-chain referenda and decided by validators, with approval from two-thirds of the active validator set, minimum participation, and a 14-day enactment delay. A team FAQ updated October 20, 2026, instead describes token-holder referenda using Substrate OpenGov conviction voting, with a 67% approval threshold, a support floor, and the same 14-day timelock. An older governance description also refers to validator voting on FIPs.^[1]​^[9]​

Both governance accounts describe an initial period in which only the FLOP Foundation may submit protocol-upgrade proposals until the first halving, though it cannot enact an upgrade unilaterally. Hayes has called the interim arrangement a “constitutional monarchy.” The whitepaper draft says any party may submit proposals after that transition, while the team FAQ says the community selects proposals for validators to consider.^[1]​^[9]​

The chain is specified to launch with a superuser key held by a multisignature wallet. A one-way handover intended to remove the key and route upgrades through governance had been designed but was not yet completed according to the draft and team FAQ. The whitepaper also describes time-locked emergency overrides and module pauses as circuit breakers, with mandatory delays and cooldowns.^[1]​^[9]​

Agent Services and Compute Markets

Agents are intended to use self-generated decentralised identifiers (DIDs) and cryptographic signatures to make requests and authorize spending. The whitepaper lists support for agent and payment standards including Agent2Agent (A2A), Virtuals ACP job vocabularies, x402, and Model Context Protocol servers. It also describes a native wallet and planned ways for agents that do not hold $FLOP to access compute through brokers accepting fiat or stablecoins. These integrations and broker arrangements are design plans, not reported network operations.^[1]

Technocore is a separate commerce and coordination platform described as part of the project’s ecosystem. It provides rooms, task and service listings, shared notes, and agent memory. The whitepaper describes the Technocore Lock Protocol (TCLK) as a way for agents or groups to record agreements and cryptographic completion receipts; it distinguishes this coordination function from the FLOP chain’s role in currency and settlement. FLOP escrow between agents and cross-chain settlement were planned for introduction with the chain.^[1]

The network’s proposed settlement primitive is a hash time-locked contract (HTLC), in which funds are locked under a cryptographic condition and deadline. The whitepaper also describes point time-locked contracts (PTLCs), which use a curve point and corresponding secret rather than a shared on-chain hash. It cautions that cross-chain settlement depends on timing, inclusion, participant binding, and finality on each chain, and does not provide an unconditional completion guarantee. At the draft’s stated status, the hash-lock path had been tested, while the point-lock path was unaudited reference cryptography and no deployed chain pair had been qualified.^[1]

FLOP Compute Units (FCUs) are a planned forward-pricing instrument for inference capacity. An FCU represents a miner-bonded commitment to provide a fixed rate of compute during a prepaid future window of up to 30 days, at a price set when the reservation is formed. Buyers are to redeem the reservation by opening ordinary inference sessions; miners must serve the reserved capacity or risk losing collateral. The design is take-or-pay, so unused capacity is not refunded, and the draft says an FCU can be transferred or divided. FCUs do not cover model training.^[1][6]

Hayes separately proposed using compute markets as the basis for a reference spot price and financial derivatives. His essay describes an hourly, volume-weighted Flop Reference Compute Rate and proposed futures and perpetual swaps tied to it, with FCUs serving as a way to construct forward exposure. These derivatives were presented as a proposal, not as confirmed operating features of the network. Flop Labs later characterized Hayes’s compute-derivatives proposal as his third major design, following the perpetual swap and a synthetic-dollar idea that Ethena used as the basis for USDe.^[6][7]