What is Pearl (PRL)?

Pearl (PRL) is a proof-of-work cryptocurrency mined by multiplying matrices on graphics cards, the same arithmetic used to run AI models. The network launched on 27 Apr 2026. About 340 million of its fixed 2.1 billion PRL supply had been mined by 10 Oct 2026, although an independent study found that most mining produced no AI output.

What is Pearl (PRL) Coin?

Pearl runs a standalone blockchain with its own coin, PRL. It follows Bitcoin's design in every respect except the work miners perform. The core software was forked from Bitcoin software called btcd, btcwallet and neutrino.

Under proof of work, computers spend computing power competing to add the next block of transactions. The winner receives newly issued coins. Bitcoin uses hashing, a guessing game with no other output. Pearl instead uses matrix multiplication, which multiplies large grids of numbers.

Matrix multiplication is the main cost of running AI. The research paper behind Pearl estimates that it accounts for at least 50% of the computing load in neural network training, and often 70 to 80%.

Pearl's figures come from the original Pearl whitepaper.

How Pearl's proof of useful work works

Miners choose two matrices and multiply them. The protocol then uses pieces of the answer as lottery tickets for the next block. When the work serves an AI model, those matrices are usually the model's stored weights and the data passing through it. The whitepaper sets out six steps.

1. Commit: The miner publishes a fingerprint (a hash) of both matrices, tied to the latest block. This prevents old work from being reused or the inputs from being changed later.
2. Add noise: Random numbers generated from the fingerprint are added to both matrices. Even an easy input, such as a matrix full of zeros, becomes as hard to multiply as any other.
3. Multiply: The GPU multiplies the two noised matrices. This is the expensive part, using the same operation an AI model would run.
4. Check the tickets: The output is divided into small blocks called tiles. Each tile wins if its hash falls below the network's target. More arithmetic produces more tickets.
5. Prove it privately: A winning miner packages the result in a zero-knowledge proof. It confirms that the calculation was performed correctly while keeping model weights and user data hidden.
6. Remove the noise: Subtracting the noise recovers the true product. The noise's compact mathematical structure makes this step cheap.

The authors describe the additional cost of the mining proof as almost negligible, writing it as "1 + o(1) multiplicative overhead". Their security argument relies on a conjecture, an unproven mathematical assumption that nobody can compute the noised product faster than by performing the full multiplication.

For privacy, Pearl uses Plonky2, a proof system built only on hash functions. Three rounds of compression reduce each block's proof to below 60 kilobytes, according to the whitepaper, within a 65-kilobyte limit. Zcash uses the same family of proofs to conceal payment details and is covered in our Zcash network statistics.

Mining difficulty, or how hard it is to find a block, adjusts after every block. Pearl accepts only Taproot, Bitcoin's newest address format, and its wallets carry a dormant backup key for XMSS, a signature scheme designed to resist quantum computers.

PRL supply, emissions and launch distribution

Block rewards created every PRL in existence. Each reward is slightly smaller than the last. The whitepaper fixes total supply at 2.1 billion PRL. Cumulative issuance at block t follows the formula 2.1 billion × t ÷ (t + 650,226).

Using the block number displayed on Pearl's block explorer on 10 Oct 2026 produces the following figures.

At the target block pace, half the supply will have been issued around the end of 2029. The first block paid about 3,230 PRL. Rewards have therefore fallen roughly 30% in under six months.

The issuance formula contains no investor or team allocation, so PRL has none of the vesting dates listed on our token unlock calendar. Its launch distribution was nevertheless uneven. The explorer places block 40,000 on 3 May 2026, just six days after launch. At the target interval of 194 seconds, those blocks should have taken about 90 days.

Early miners received about 121.7 million PRL from those blocks, equivalent to 36% of all coins mined to date and 5.8% of the final supply. The whitepaper says difficulty started low to give early miners a fair opportunity and to "ensure the core team has minimal resources needed to support the massive research, development and infrastructure" the project requires. The team has not published its holdings.

Who built Pearl (PRL)

Pearl Research Labs is led by an academic computer scientist, so the project's credibility rests on published research. Together AI's partnership announcement names Omri Weinstein as co-founder and chief executive. He is an associate professor of computer science at the Hebrew University of Jerusalem.

According to his Simons Foundation profile, Weinstein completed his PhD at Princeton in 2015 before joining Columbia University's theoretical computer science group. He co-authored Pearl's founding paper with cryptographer Ilan Komargodski in April 2025.

Komargodski also co-authored Hawkeye with Stanford cryptographer Dan Boneh. That paper shows how an ordinary processor can reproduce a GPU's exact arithmetic. In June 2026, cryptographer Rafael Pass published an economic model of proof of useful work.

When we checked Pearl Research Labs' website on 10 Oct 2026, it listed no investors. The site displays an Nvidia Inception startup program badge. The company also sells AI inference, the process of running a trained model to produce answers, as a commercial service.

Is Pearl's useful work useful?

The protocol verifies matrix multiplication without checking whether anyone wanted the answer. Its founding paper lets miners choose their own inputs and defines usefulness by whether someone will pay for the result.

Abhinaba Basu examined miners' behaviour in a preprint titled The Usefulness Gap in Proof-of-Useful-Work, published in June 2026.

• Random inputs were accepted: Basu built a miner that multiplied random matrices without an AI model. A mining pool, whose members share rewards, accepted 44 of its shares.
• The main mining software ran no AI: His analysis covered 8,012 machines run by the top 15 miners in AlphaPool, a pool with about 21% of the network's mining power. Based on string analysis, the dominant mining program appeared to contain no code for running models.
• The energy cost was large: He estimated that the network used the equivalent of 320,000 RTX 3090 graphics cards drawing 112 megawatts.
• GPU rental prices rose: On one marketplace, budget GPU rental prices increased 38% after public mining software became available. Basu attributed 23 percentage points of that increase to Pearl.

Running a real model requires time and memory. Basu estimated that loading one would reduce mining output by 10 to 30%, leaving a miner paid only in PRL better off skipping it.

Pearl's commercial evidence comes from Together AI. On 15 May 2026, the company launched a version of Google's Gemma 4 model running on Pearl's protocol at more than 25% below its usual price. Together AI said the discount is "offset by the future value of crypto emissions" earned by those same GPUs.

Basu describes this as a financial subsidy funded by mining income. Paid AI work runs on Together AI's machines, he argues, while the wider mining network contributes nothing to it. Pearl's whitepaper acknowledges part of that criticism, stating that the network "attracts compute that does not necessarily have useful work".

Pass's model finds that block rewards can provide a rebate on AI prices once enough paid work is mined. The cost of attacking the chain remains tied to the block reward. Reaching that state would require paid AI work to become the cheapest way to mine Pearl.

Recent changes to Pearl mining and trading

Changes to Pearl's rules, hardware requirements and trading venues affect who earns PRL.

• A move to floating-point mining: Pearl launched with whole-number arithmetic. Production AI models increasingly use FP8, an 8-bit number format that exchanges a little precision for speed. A specification published in September 2026 describes mining directly on FP8 multiplications.
• A narrower hardware list: PIP-3, the proposal to activate that change, approves two Nvidia chip families, Hopper and Blackwell. It remains a draft with no activation block. On 7 Oct 2026, mining pool Kryptex wrote that older consumer cards could not produce valid proofs under the draft as written.
• Frequent rule changes: A hard fork is an upgrade that every computer on the network must adopt, while a soft fork only tightens the existing rules. A hard fork added support for mixture of experts, a type of AI model, at block 71,935. A soft fork at block 91,630 then rejected mixture-of-experts proofs again. Another soft fork at block 96,251 set a minimum size for the added noise and removed the easier target that larger noise had given miners. A hard fork at block 99,000 changed how the noise is generated. PIP-3 would require another hard fork.
• A derivatives market in preparation: On 26 Sep 2026, Entropy paid 500 HYPE, Hyperliquid's token, worth about $45,000, for the PRL ticker, according to Lookonchain. Entropy lists markets on Hyperliquid, and the purchase points to a planned PRL perpetual futures market. These contracts track a price without an expiry date. On 10 Oct 2026, Hyperliquid's market data showed the PRL market set up but marked as delisted. Nothing had traded on it. Our Hyperliquid statistics cover the exchange.

In a 10 Oct 2026 report, Chinese-language outlet TechFlow placed Pearl among the most discussed projects at the Token2049 conference. Supporters called it the Bitcoin of AI, while critics cited the usefulness gap.

Pearl Research Labs' direct inference service listed seven open models from Qwen, Z.AI, DeepSeek and Google on its pricing page on 10 Oct 2026. Customers pay in US dollars per unit of text processed, so the service creates no direct demand for PRL.

Where PRL trades and what market data shows

A handful of small exchanges list PRL. Its price reflects expectations far more than current use. CoinGecko's Pearl page reported these figures at 09:00 UTC on 10 Oct 2026.

Fully diluted valuation multiplies the price by all 2.1 billion PRL that will ever exist. Market capitalisation uses the 340.3 million PRL in circulation, which matches the figure the issuance formula gives.

• Two venues hold almost all trading: SafeTrade's PRL/USDT pair accounted for 66.8% of reported volume and BigONE's PRL/USDT pair for 32.7%.
• New supply is large against trading: At that price, daily issuance of about 1.01 million PRL was worth roughly $1.41 million, equal to 32% of the day's trading volume.
• PRL pays holders nothing: Holders receive no staking rewards for locking up coins. PRL also offers no fee sharing or governance vote. Our crypto project revenue dashboard tracks protocols that earn fees, but PRL has no equivalent income to measure.

The whitepaper describes the coin's store-of-value role as "completely dependent on the expectation of its future adoption as a means of payment".

Because PRL runs on its own blockchain, it has no contract address. Perle, a separate token on Solana, shares its PRL ticker. A PRL pair on a decentralised exchange, including one found through our DEX screener, is therefore a different asset. Check both the project name and ticker on any price list, including our live crypto prices.

The risks of holding or mining PRL

PRL is a young coin with heavy issuance, and its main claim remains unproven on its own network. Holders and miners face several ways to lose money.

• Unproven usefulness: The network measured in June 2026 produced no AI output. Pearl's case for wasting less energy than Bitcoin depends on attracting paid AI work.
• Concentrated early supply: More than a third of mined PRL was issued at low difficulty within six days. Those coins have no lock-up, allowing their owners to sell at any time.
• Steady dilution: About 1.01 million new PRL enter the market each day. The price falls unless buyers absorb that supply.
• Few venues: Two small exchanges handle almost all reported volume, and SafeTrade alone handles two-thirds. A large order can move the price sharply, and withdrawals depend on those exchanges remaining solvent.
• Rule and hardware changes: Pearl Research Labs writes the upgrades. Miners buying or renting consumer GPUs could find their hardware excluded if PIP-3 activates in its current form.
• Falling mining income: One widely cited profitability tracker cut its daily revenue estimate for an RTX 5090 from about $33.80 to $17.19 within weeks as more miners joined, Hashrate Index reported in June 2026.
• Leverage: If PRL perpetual futures become available, positions opened with borrowed funds can lose the entire amount posted as margin, the deposit backing the trade.