¶ PEARL is the currency of AI. Daily stats, updates, and insights.
Tag us on Pearl content
Joined May 2026
- Tweets138
- Following5
- Followers462
- Likes240
The classic worry for any proof-of-work chain: control most of the mining power and you can rewrite recent history.
Rafael Pass’s paper shows the cost of a majority attack on pearl-2:native is set by the block reward, and is at least as high as under Bitcoin-style proof of work at the same token-to-inference ratio, even though Pearl’s miners also earn from AI work.
“When FP8?” comes up in most pearl-2:native threads.
Where it stands on Oct 28:
The FP8 upgrade (certificate V4, a hard fork) sits in pull request #311, open since Sep 9, with no mainnet block height in the code.
On Sep 25 @prlnet CEO said it was on testnet and expected to launch “within weeks”. Work on its proof system is still in progress, with prover commits as recent as Oct 25. The signal to watch is a fork height for mainnet.
Pearl mainnet is six months old today.
Block 1 was mined on Apr 27 at 09:00 UTC.
Six months on, the chain is at block 134,576 with 17.15% of the 2.1B pearl-2:native cap mined. The network hashrate high stands at 86.08 EH/s, set Oct 23.
Pearl Research Labs @prlnet made the front page of @TheMarker, the reputed Israeli business daily, on Oct 25.
In an exclusive, Ofir Dor @doritsos reports that the Tel Aviv lab raised $130M without announcing it. Pearl hasn’t announced a funding round on its own channels.
The piece covers Pearl’s pitch: cut the cost of AI by paying for inference compute in pearl-2:native. Its caveat is that the economics haven’t yet been proven at scale.
pearl-2:native 's supply passed 17% of its 2.1B cap at block 133,179 on Oct 25, two days before the network turns six months old.
By the emission formula, half the cap is mined at block 650,226, about 517,000 blocks from here, or roughly 3.2 years at the 194-second block target.
New high for pearl-2:native’s network hashrate: 86.08 EH/s, set Oct 23.
Difficulty reached 44.0M, up 3.6% in about 20 hours.
Pearl Research Labs @prlnet published PearlQuant: a method for compressing the KV cache that LLM serving keeps in GPU memory.
On GLM-5.3 it stores 3.2 bits per value against 9.1 for the FP8 baseline, about 65% less, with no statistically significant difference on four benchmarks (RULER, LongBench v2, MMLU-Pro, LiveCodeBench).
On one 8×B200 node at 128 concurrent sessions it gave 1.8x FP8’s throughput and cut p90 time to first token from 144 s to 7 s. At light load, FP8 was about 5% faster.
Source: pearlresearch.ai/research/bl…
New high for Pearl’s network hashrate: 85.62 EH/s.
The previous high was 79.15 EH/s on Oct 20.
pearl-2:native blocks averaged 138.5 seconds over the last 60, against the protocol’s 194-second target.
Difficulty sits at 42.7M, up 19% in a week (36.0M on Oct 15).
Every pearl-2:native in existence was mined from the genesis block on April 27, 2026. The team kept nothing back for itself and there was no private sale. That’s what a fair launch means.
New high for Pearl’s network hashrate: 79.15 EH/s.
The previous high was 70.57 EH/s on Oct 18.
pearl-2:native blocks averaged 150.4 seconds over the last 60, against the protocol’s 194-second target.
Difficulty sits at 40.6M, up 17% in a week (34.7M on Oct 13) and 22% since Oct 3 (33.2M). It adjusts on a weighted average with a one-week half-life, so it trails a fast climb in hashrate.
Pearl node v1.5.0 is out (Oct 13), a security release: hardening across P2P, RPC, SPV, the mempool and the wallet, for issues reported through GitHub security advisories.
Twelve researchers and Zellic are credited.
No fork height changed, and the FP8 pull request is still open.
Pearl node v1.5.1 is out (Oct 15), a consensus fix. A transaction could spend an output created by a later transaction in the same block.
Nodes now require parents before children inside a block, and re-check blocks that older versions had marked valid. No fork height changed. The PR credits the researcher who reported it.
Pearl node v1.5.2 is out (Oct 17), the third node release in five days.
It fixes how the node reads its UTXO cache: the gettxout RPC could read an entry after releasing the chain lock while a new block was changing it. The fix credits an outside researcher and Kryptex, the pool that found the most blocks this week. No fork height changed.
Animation explaining Pearl’s Proof of Useful Work, in which matrix multiplications performed on GPUs for AI inference also generate verifiable work for network security and $PEARL issuance.
Documenting ¶Pearl retweeted
The math under Pearl's 2-for-1 comes from a 2025 paper by Ilan Komargodski and Omri Weinstein: proof of work on arbitrary matrix multiplication.
The miner picks both matrices, so they can come from a model the GPU is already serving. The paper proves the extra cost shrinks as the matrices grow. Today's INT8 mining adds about 4-15% depending on workload, and the FP8 upgrade targets 0-3%.
The operation that runs AI can also mint $PRL.
