Artificial intelligence needs computing power. Today, that power is usually bought from large cloud providers. FLOP Network is a proposed blockchain network designed around a different idea: AI agents should be able to buy computing power directly from independent providers, while the network checks that the requested AI work was actually performed.
The concept is ambitious, technical, and still under development. This guide separates the project’s stated design from what readers can actually do today. It is intended for curious newcomers, crypto users, developers, and people considering a future testnet.
Important: This article relies on FLOP’s own website and technical documentation. The project’s public documents are drafts in several places, and the official Teaser explicitly says that figures may change. Nothing here is financial advice, a promise of access, or a guarantee of an airdrop.
Arthur Hayes and the people publicly identified behind FLOP
Arthur Hayes is the most prominent public figure associated with FLOP. He is best known in crypto as a co-founder and former CEO of BitMEX, and describes himself as Chief Investment Officer of Maelstrom. In his own August 2026 essay introducing FLOP’s origin story, Hayes presents the network’s central idea, says that he self-funded the team building it, and directs readers to FLOP Labs for further information.
Primary Source:
Arthur Hayes introduced FLOP in "The Book of Genesis" on his Substack
and Arthur Hayes profile -- access on Substack required
That association is relevant context, but it is not a substitute for evaluating the protocol, its code, or its future launch conditions. At the time this article was checked, FLOP Labs’ public website and public GitHub organisation did not publish a named roster of additional founders, executives, investors, or technical leads. It would therefore be misleading to present other people as being “behind FLOP” without a direct announcement from FLOP Labs or from the person concerned. This section will be updated only if primary sources identify them.
The short version
FLOP is designed as a blockchain for useful AI inference.
Inference is the moment an already-trained AI model receives an input and produces an output, for example, when a language model answers a question.
In the intended system:
- An agent asks for an inference job and offers a fee in FLOP.
- A miner, in this context a compute provider, not a hash miner, runs the job on suitable hardware.
- The miner supplies evidence that the work was performed.
- Validators record and verify the relevant evidence, then settle payment and rewards through the network.
FLOP is therefore meant to be both the network’s token and the unit used to purchase inference. The project calls this model Proof of Useful Inference (PoUI). FLOP overview
Why does FLOP exist?
Most blockchains reward computers for securing a ledger. FLOP’s proposal is to reward computers for performing an economically useful task: AI inference that someone has requested and paid for.
That creates a difficult problem. It is easy to claim that an AI model ran; it is much harder to establish that it ran correctly without simply repeating the same expensive computation. FLOP’s design attempts to make false claims costly and detectable through several layers, including execution traces or commitments, selective re-execution, validator checks, and stake that can be penalised. The detailed design is in the project’s Yellow Paper.
For a non-technical analogy: imagine paying a delivery company. A receipt alone is not enough; you want tracking, proof of delivery, and consequences for dishonest claims. FLOP is trying to create an equivalent set of checks for AI compute.
The three roles, in plain English
1. Agents: the users of compute
An agent is the party that needs an AI task performed. It sends a request that is designed to specify key conditions such as the model, maximum latency, expected compute, a confidentiality choice, and a fee. The intended result is a system where an agent can convert FLOP into inference on demand. Agent documentation
For most people, this is the easiest role to understand: it is the buyer of AI compute, rather than the operator of infrastructure.
2. Miners: the providers of compute
In FLOP, a miner is a provider that runs inference workloads for agents. It is not the traditional proof-of-work activity of guessing hashes. A miner is expected to operate suitable GPU hardware, accept jobs, perform them, and provide the required evidence.
The current design says that miners stake FLOP and can be penalised for dishonesty. It also describes a default SOFT route for ordinary capable GPUs and an optional HARD route involving confidential-computing hardware. The public miner page also states a proposed minimum base self-stake of 10,000 FLOP plus a capacity-linked requirement. These are protocol-design details, not a present invitation to buy hardware or an assurance of profitability. Miner documentation
3. Validators: the network’s security operators
Validators are expected to operate the chain’s consensus and verification infrastructure. In the proposed design, they build blocks containing commitments to inference proofs, support data availability for model weights and evidence, and participate in governance.
This is an infrastructure and staking role, not a GPU-serving role. It carries operational and economic risk, including possible slashing for dishonest behaviour. Validator documentation
How a FLOP inference session is intended to work
The public documentation describes a simple flow:
Agent requests an AI task and offers a FLOP fee ↓ Miner accepts the task and runs the inference ↓ Miner returns the result and proof-related evidence ↓ Validators record the relevant proof commitment and settle the session
The key idea is that payment should be tied to verifiable work, rather than trust alone. A request is designed to include a model-weight hash, latency target, compute amount, confidentiality flag, and fee. If the result is disputed, the network design includes a challenge and adjudication path. FLOP overview
This is a design goal, not a claim that every feature is already available to public users.
What is officially known about the roadmap
As of this article’s date, FLOP’s public Teaser targets a Q4 2026 testnet and Q1 2027 mainnet. In an official 6 October update, FLOP Labs said the testnet is expected in late October or early November and should run for around 90 days. The Teaser also says that its figures are provisional and that the Yellow Paper is presented as the definitive specification, although it is not yet final. These are targets, not launch guarantees: software, documentation, eligibility rules, and infrastructure details can change before release. FLOP Teaser
That distinction matters. At the time of writing, readers should not assume that a wallet, chain identifier, RPC endpoint, faucet procedure, public explorer, hardware checklist, or final client workflow has been announced unless it appears in an official FLOP publication.
Testnet and potential airdrop: facts versus assumptions
This is the section where caution matters most.
FLOP’s 30 September 2026 draft update states that testnet participation is intended to determine the genesis airdrop across agent, miner, and validator roles. For agents, the current description is a future test-token faucet and spending those test tokens on inference. Any resulting allocation would initially be locked and usable only for compute; the schedule for becoming liquid has not been set.
However, this does not mean that every person who signs up, owns a GPU, creates an account, posts online, or follows a social account will receive tokens. The Teaser says its figures are provisional, and the Yellow Paper contains design parameters that remain subject to finalisation. Eligibility rules, anti-abuse controls, allocation thresholds, operational requirements, and conversion mechanics should be treated as unknown until FLOP publishes them directly.
The honest takeaway is simple: FLOP now states that testnet participation is intended to determine the genesis airdrop, but the details remain provisional.
- There is no responsible basis for promising an allocation to any individual.
- Airdrop farming should never outweigh security, cost control, or genuine use of the network.
How to prepare responsibly
Preparation should reduce the risk of missing an official launch, not create unnecessary risk or expense.
Follow primary sources
Use FLOP’s website, its official social channels, and the project’s own repositories or documentation when they are published. Treat screenshots, referral threads, imitation tokens, and claims that require payment or private keys with suspicion.
Learn the role before choosing it
The agent role is conceptually the most accessible because it is about requesting inference. Mining and validating may require capital, hardware, staking, reliable operations, and an ability to absorb technical or economic losses. Do not rent GPUs or buy equipment based solely on an expected reward.
Keep a dedicated security setup
When official FLOP software is available, use a dedicated wallet or account for the project, protect recovery material offline, and verify domains carefully. Never share a seed phrase, private key, or password in exchange for a supposed whitelist, faucet, allocation, or technical support.
Wait for official technical instructions
It is reasonable to read about GPU capacity, nodes, or accounts. It is not reasonable to invent a setup before FLOP publishes the relevant network information. In particular, do not connect a wallet to an unverified site or send funds to “activate” an allocation.
Contribute because the contribution is useful
Good contributions include accurate documentation, testing when officially opened, bug reports through the project’s stated channels, and educational material that distinguishes fact from speculation. Repetitive posts, false claims, sybil activity, or fabricated engagement can harm readers and may be filtered by any future anti-abuse system.
A practical checklist
Before the testnet:
- Read the FLOP overview and the Yellow Paper.
- Follow official FLOP announcements and enable notifications if appropriate.
- Decide which role you understand and can afford to operate safely.
- Keep a record of official URLs; do not rely on search ads or forwarded links.
- Do not spend money on hardware, tokens, or access solely because of an airdrop rumour.
When official testnet instructions arrive:
- Confirm the announcement through an official FLOP domain or official account.
- Read the rules in full, including eligibility and anti-abuse terms.
- Use only the documented network settings, software, and links.
- Start small and record what you do.
- Report issues honestly through the official channel.
What FLOP is not, at least not yet
FLOP should not be confused with a guarantee of cheap AI compute, a guaranteed passive-income machine, or a confirmed airdrop for early followers. It is a proposed protocol with a detailed technical direction and an announced testnet target. Its success will depend on execution: useful demand from agents, reliable compute supply, robust verification, sound incentives, and transparent launch rules.
For now, the most sensible stance is informed curiosity. Understand the design, monitor primary sources, protect your keys, and take action only when the official path is published.
Sources
All factual protocol claims in this article are based on primary FLOP sources. The biographical context on Arthur Hayes is sourced from his own publication:
- FLOP Network overview
- FLOP Yellow Paper
- FLOP Teaser and roadmap
- FLOP agent documentation
- FLOP miner documentation
- FLOP validator documentation
- Arthur Hayes, The Book of Genesis
Last updated: 6 October 2026. FLOP’s documents remain drafts. A previous agent-unlock ratio was removed because it no longer appears in official documentation. FLOP Labs now expects the testnet in late October or early November, for around 90 days.
Source: https://x.com/flop_labs/status/2107376810472378619