Pearl (PEARL): $1.6M a Day of New Coins Pays for "Useful" Work Nobody Ordered

Score: Legit chain, bogus headline. Pearl is a real open-source blockchain whose mining math checks out — we verified its issuance against the chain itself and found no premine hiding in it. But the pitch on the tin, "proof of useful work," describes what the protocol allows, not what it measures. The chain pays out 1.06 million new PEARL a day — about $1.6 million at the current price — for matrix multiplications it cannot distinguish from random numbers nobody asked for, and an independent measurement found exactly zero useful AI computation happening. The payer isn't AI customers. The payer is inflation. Here is the math.

The Pearl logo. Logo: Pearl Research Labs, shown for editorial identification.

Official website: pearlresearch.ai

What Pearl is

Pearl is a Layer 1 blockchain launched on April 27, 2026, forked from btcd — a Bitcoin implementation written in Go. It keeps Bitcoin's skeleton: UTXO accounting, Taproot scripting, a hard supply cap, and mining as the only way new coins enter circulation. It changes exactly one thing: the work.

Bitcoin miners burn electricity on SHA-256 hashes with no value outside the protocol. In Pearl, the proof of work is a matrix multiplication — the operation at the heart of every AI model. The miner picks its own matrices, the protocol perturbs them with low-rank noise derived from the chain state (so the multiplication can't be shortcut), the GPU multiplies them in FP8, and the output is cut into tiles hashed with BLAKE3. Any tile under the difficulty target wins the block. A zero-knowledge proof (Plonky2) lets the network verify the multiplication happened without seeing the inputs.

The pitch writes itself: if the matrices happen to be a real AI workload, the same GPU cycles produce intelligence and consensus — a "2-for-1." Pearl's kernel drops into standard inference stacks; team benchmarks put the mining overhead at ~4–5% when real inference runs through it, other measurements at 10–15%. The problem is in the next two sections.

First, the basics: PEARL is a coin, not a token

Every review on this site states this up front: PEARL is a coin. It has its own blockchain and its own miners, and no contract address anywhere. Addresses start prl1p, and the smallest unit is a grain (10⁻⁸ of a coin).

Two naming traps. The project's own materials mostly use the ticker PRL; CoinGecko and newer listings use PEARL — same coin. Two unrelated projects also use PRL: Perle, a Solana token, and Oyster Pearl, a dead 2017–2018 token whose founder exit-scammed and was charged by the SEC. Neither has anything to do with this Pearl. Check the chain, not just the ticker.

The usefulness gap: the chain can't tell a customer from a coin flip

Here is the design decision everything hangs on. Pearl verifies that a matrix multiplication was performed correctly. It does not — by design cannot — verify that the multiplication was worth doing. The miner chooses its own inputs; nothing checks whether the matrices came from a real AI model, whether anyone requested the computation, or whether the result reached anyone. Random matrices pass every check a paying customer's matrices pass.

This isn't a theoretical loophole. It's the measured state of the network:

• The one systematic measurement found zero. A June 2026 empirical study (Basu, arXiv:2606.04819 — a preprint, still under review, and we label it as such) examined 8,012 workers on a major pool over May 30–31, 2026 (~a fifth of the network). The dominant mining software contained no inference code at all. The author's own random-matrix miner had 44 shares accepted by a pool. His conclusion: the network "produces zero useful AI computation."
• The team doesn't dispute the mechanism. Co-founder Omri Weinstein has publicly acknowledged that as long as miners choose their own computation, useless mining cannot be prevented. PANews found most hashrate running on random data — "essentially no different from Bitcoin mining."
• The fix exists only as an announcement. In September 2026 Pearl and a partner called Attestable announced a "Proof of Inference" layer to tie mining math to genuine paid jobs. As of mid-October coverage it was still in early development — not live, not measurable.

So "proof of useful work," November 2026: permitted by the protocol, pitched by the marketing — not required, not verified, and on the only independent count, not happening.

The one customer story, checked: Together AI

Pearl has one named commercial relationship — the strongest evidence for the pitch. In May 2026, Together AI launched a serverless endpoint for a Pearl-tuned model, Gemma-4-31B-it-Pearl, priced more than 25% below Together's standard rate. Pearl's own site describes it as "funded by the PRL mined on the same forward pass."

Read that sentence carefully, because it gives away the direction of the money. The discount isn't funded by customers paying miners. Per Together's own description, the lower price is offset by the expected future value of PRL emissions — new coins subsidize the API price. The customer pays Together in ordinary money; the miner is paid in inflation; the two never meet. The subsidy flows from PEARL holders to API customers — the opposite direction from "AI demand pays for mining."

And the endpoint is fading: mid-to-late October coverage found the model's page marked "Not available on Serverless API," its Hugging Face provider status flipped from Live to Error. A dedicated self-hosted endpoint reportedly still works, without the discount. Neither company has published a single usage, customer, or revenue figure for it.

Pearl Research Labs sells inference — API access to open models (DeepSeek, Qwen, GLM, Gemma) for prepaid USD credits — with no published revenue figures and no published split between mined and company-run compute. Today there is no marketplace where a buyer pays miners for inference, in PEARL or anything else — the coin captures none of the inference revenue. The "2-for-1" is a roadmap, not a cash flow.

Who actually pays: the emission math

If customers aren't paying the miners, who is? Anyone holding the coin. Pearl issues new PEARL on a smooth polynomial decay instead of Bitcoin-style halvings. The whitepaper's formula: the fraction of the 2.1 billion cap issued by block t is t / (t + H), with H = 650,226 blocks — about four years at target pace, the halfway point of all issuance. We didn't take the project's word for it. We read it off the chain (Blockbook explorer backend, November 1, 2026):

	Verified figure
Genesis block	April 27, 2026, 09:00 UTC — coinbase paid 0 (no premine in block zero)
Block 1 reward	3,229.64 PEARL
Block 137,043 reward (Nov 1, 2026)	2,203.12 PEARL — exactly what the formula predicts
Blocks per day (recent pace, ~180s vs 194s target)	~480
New PEARL issued per day	~1.06 million
Value of that issuance at $1.53	~$1.62 million per day
Circulating supply	365.5M PEARL — 17.4% of the 2.1B cap
Market cap / fully diluted value	$560.6M / $3.22B

Three takeaways. First, the schedule is honest: the formula-implied issued total (365.55M) matches CoinGecko's circulating figure (365.5M) to four significant figures, and every coinbase we sampled pays one miner address with no developer fee skimmed. Second, the dilution is ferocious — at today's pace the chain issues new coins equal to roughly 106% of the entire current circulating supply per year. PANews calculates ~20% of total supply released in year one against Bitcoin's 12.5%; Blockworks' bear case puts dilution near 85% this year. Third, this is the entire miner wage and security budget, paid by holders — the same payer classification as Helium and most DePIN: emissions, not customers.

Mining Pearl today: the best-paid GPU job in crypto

Pearl is currently the most profitable thing a commodity GPU can do. Checked November 1, 2026:

• Network hashrate: ~79 EH/s in the trackers' hash-equivalent units (PearlTrack: 79.26 EH/s), down from a peak near 101 EH/s on October 27.
• An RTX 4090 earns roughly 3.7–4.2 PEARL a day depending on calculator and tune (WhatToMine: 330 TH/s at 370W → 4.09 PEARL, $6.22 revenue, $5.33 profit at $0.10/kWh; hashrate.no and Kryptex land in the same band). MinerCompare puts Pearl at 3–4× the profit of the best established GPU coin on the same hardware — the subsidy talking.
• The wage is already collapsing. Revenue per unit of hashrate is down about 84% from its late-May peak as GPUs piled in; per-card gross peaked at $11.40 on October 21 and fell by a third within days. PANews found GPU renters underwater by late October — renting a 4090 on Vast.ai cost ~$8.69 a day against ~$7.36 gross. Own the card and cheap power, and Pearl pays. Rent either, and it doesn't.
• GPU-only in practice, overwhelmingly NVIDIA — community miners (SRBMiner, BzMiner, Kryptex's KRig) cover RTX 20- through 50-series and some AMD. The opposite end of mining from Monero, where a CPU earns cents: Pearl is the subsidy the GPU fleet Ethereum set free in 2022 has been waiting for.
• Pool concentration is the quiet problem. PearlTrack's audited table for November 1: Kryptex 48.1% of blocks, PearlHash 18.0%, Pearl Fortune 11.0%, LuckyPool 7.5%, Hero Miners 5.0%, AlphaPool 0.4%, ~10% unattributed. Two pools control roughly two-thirds of blocks — a Nakamoto coefficient of 2 — and on October 27 Kryptex alone topped 53%. PearlTrack audits claimed hashrate against what blocks prove — and the proven picture is top-heavy.

Who's behind it

The company. Pearl Research Labs builds the chain, the mining software, and the inference platform. There is no foundation, and the company hasn't disclosed its legal structure, headcount, or its own PEARL holdings (DataWallet). One company, closed books.

The founders are academics, and their records are clean. Co-founder and CEO Omri Weinstein: theoretical computer scientist, Princeton PhD, computer science professor at the Hebrew University of Jerusalem, with industry time he lists at Nvidia and Vast Data (some profiles add Microsoft). Co-founder Ilan Komargodski: cryptography professor at the Hebrew University, Weizmann PhD, research time at Cornell Tech and NTT Research. Third co-founder Idan Sugarman runs the business side; his prior employment couldn't be verified from a primary source. Cornell Tech's Rafael Pass wrote Pearl's economics paper — while consulting for the Labs, per the paper's own footnote. A skeleton check (failed startups, collapsed tokens, fraud settlements) found none. Publishing academics, not serial token launchers.

The money is reported, not confirmed. As of late October, no funding round had been officially announced. On October 25, 2026, the Israeli business daily TheMarker reported a stealth raise of roughly $130 million — ~$55M equity plus ~$75M for a separate GPU infrastructure fund — naming Lightspeed, Ribbit Capital, and Square Peg. Pearl hasn't confirmed it; neither has any named firm. We print the $130M as reported, not confirmed. One pin-downable affiliation: the NVIDIA Inception badge on Pearl's site is a membership program, announced in July — not an NVIDIA investment or partnership, whatever the logo wall implies.

Control. There are no token allocations — no team tranche, no investor tranche, nothing vesting, nothing to unlock. Control lives in three places: the team's undisclosed launch-window mining (next section), the two-pool concentration above, and the Labs' grip on the reference software. Skin in the game is unverifiable: the team says it mined early to fund development, but with no disclosed wallets and no published audit of the node or proof circuits (DataWallet found none), holding-versus-selling can't be checked.

The fair launch, with one asterisk

Pearl's code is as clean as launches get: mining is the only source of PEARL — no ICO, no presale, no private token round, no founder allocation in the code (confirmed by our own read of the chain). Genesis paid zero.

The asterisk is the first week. The whitepaper sets the initial difficulty deliberately low, and says why: a fair shot for early miners before AI-enterprise hashrate arrives, and resources for the core team. We verified the result on-chain: block 40,000 landed May 3, 2026 — 5.9 days after genesis — instead of the ~90 days those blocks take at the target pace. Those blocks minted 121.7 million PEARL, a third of today's entire circulating supply. Blockworks' bear case (October 23, 2026) puts the open question plainly: roughly a third of today's supply was mined in the first six days, "and we don't know how much of that went to the team." Neither do we. Fair in the code, front-loaded in practice — and the overlap is a number the team has never published.

The record so far: one exploit, four forks, no theft

• July 2026 — the MoE exploit. An optional mixture-of-experts extension let miners forge proofs and overstate hashrate, winning a disproportionate share of block rewards. The team killed the feature — a soft fork at block 91,630 (July 23, on-chain) ended MoE proofs. No dollar figure for the forged rewards has been published and no clawback reported. Nobody's funds were stolen; miners were overpaid by the protocol itself — a quieter loss absorbed by holders through extra issuance.
• Fork cadence. MoE hard fork at block 71,935 (June), dense-only soft fork at 91,630 (July), a rank-penalty soft fork at 96,251, and a salted-noise hard fork at 99,000 (August 11) — four consensus changes in about four months. Four in a season is a lot of trust-me upgrades for a network asking to be treated as money.
• October 2026 — security hardening. Node v1.5.0 (October 13) fixed remotely crashable inputs, RPC authentication, and credentials leaking into logs, thanking a dozen reporters including audit firm Zellic. Responsive maintenance — not a substitute for the published third-party audit of the node and proof circuits that, as of November 1, does not exist.
• Nothing worse. No confirmed theft, no drained treasury (there isn't one), no regulatory action against this Pearl. The skeletons here are an overpaying bug and an undisclosed early-mining haul — real, but a different species from the usual rug.

The market you're buying into

PEARL changed hands at $1.53 on November 1, 2026 — a $560.6M market cap at rank #98 against a $3.22B fully diluted value. That 5.7× gap is the dilution table above, priced in. Four months of listed history: first close $0.56 on June 22, 2026, a washout to $0.14 intraday on July 23, a vertical run to the $2.46 all-time high on October 21, and a slide since — down 38% from the high, about 20% in the last week.

PEARL daily closes, entire listed history (June 22 – November 1, 2026): $0.56 → $1.53 (+174%); July washout and October $2.46 spike visible. Data: CoinGecko, checked November 1, 2026.

Liquidity is the other half. All $5.4M of November 1's spot volume sat on small offshore venues — SafeTrade ($1.82M), Ourbit ($1.51M), BigONE ($1.37M), Gate ($0.71M) per CoinGecko's tickers; in October PANews counted 98.8% of volume on just the first three. No major exchange lists PEARL spot. Coinbase put it on its listing roadmap on October 19 — a roadmap is not a listing and has no date; Kraken's roadmap and BitGo custody followed. Until then: a thin book where daily emission alone (~$1.6M of new coins) is roughly 30% of daily volume.

The score, revisited

The chain: legit. Published cryptography, open code, issuance verifiable to the grain, no premine in the code, no theft on record, named academics with clean histories. The headline: bogus, so far. "Proof of useful work" currently proves work, full stop — the usefulness is an unenforced option miners have measurably not taken, the one commercial inference deal is subsidized by the emissions it was supposed to replace and is marked unavailable, and the $1.6M-a-day wage bill lands on holders as ~106%-a-year dilution. Even security is circular: Pass's own model ties attack cost to the block reward, so the chain is exactly as secure as its inflation is valuable.

What would flip this score is specific and checkable: the Proof of Inference layer shipping with a published, verifiable share of hashrate tied to paid jobs — or inference revenue, from any source, routed to the coin. Until that number exists, PEARL is a well-engineered emissions machine with an AI story it hasn't proven yet — and the emissions, unlike the story, are very, very real.

Get posts like this by email: Subscribe to BogusAPY

Follow BogusAPY on X: @BogusAPY

Check it yourself: links and sources

• Official website: pearlresearch.ai — the network page carries the protocol FAQ and issuance formula
• Whitepaper: Pearl_Whitepaper.pdf (FP8 protocol spec, September 2026) and the INT whitepaper page — §5.4 is the emissions section
• Code: github.com/pearl-research-labs/pearl — node, wallet, ZK proving system, vLLM miner
• Block explorers: explorer.pearlresearch.ai (official) · pearltrack.io — Pools page (claimed vs proven hashrate audit) and Top accounts page · prlscan.com
• Rich list: Pearl is a UTXO coin, so there's no token holder chart — PearlTrack's Top accounts page (above) is the equivalent
• CoinGecko: PEARL page — the id is pearl-2; the unrelated Perle token is a different listing
• The critical read: Basu, The Usefulness Gap in Proof-of-Useful-Work (arXiv:2606.04819) — a preprint, and the only empirical measurement of the usefulness claim anyone has published
• Official socials: @prlnet on X · community: Discord · inference platform docs at platform.pearlresearch.ai/docs

Key sources, all checked November 1, 2026: the Pearl chain itself via the Blockbook backend (genesis through block-137,043 coinbases and timestamps); Pearl's official site and network FAQ (mechanism, issuance formula, Together endpoint description); the INT whitepaper §5.4 (2.1B cap, H = 650,226 blocks, 194s target, low initial difficulty); CoinGecko (price, market cap, FDV, circulating supply, ATH/ATL, tickers, daily closes); PearlTrack (hashrate, block pace, pool shares, hashrate audit); WhatToMine, hashrate.no, Kryptex, MinerCompare (GPU earnings); Basu (arXiv:2606.04819) for the usefulness measurement; PANews and TokenPost (random-matrix hashrate, Together endpoint status, Coinbase roadmap); DataWallet and PearlCoinGuide (no-premine code read, fork log, no published audit, company structure); Blockworks Research's bear case (Oct 23, 2026); TheMarker's reported $130M raise (Oct 25–26, 2026) — reported, not confirmed. Full figure-by-figure sourcing is in this package's sources file.

Not financial advice.