←
Documenting ¶Pearl175 posts
Documenting ¶Pearl avatar

Documenting ¶Pearl

@DocumentingPRL

¶ PEARL is the currency of AI. Daily stats, updates, and insights.

◷ Joined May 2026▧ 93 Photos and videos
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Jun 20

Pearl is the first money produced by AI and secured by AI. Every GPU cycle that serves an AI model can mint $PRL at the same time, for little to no extra cost. This account documents @prlnet and the rise of the 2-for-1 kernel economy. Start here ↓

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·5h

When quantum computers can break today's signatures, most chains will scramble to move old keys to safety. Pearl's consensus has accepted a hash-based signature, XMSS, and P2MR outputs from genesis, co-founder Ilan @komargodski says, so $PEARL could be locked quantum-safely from the first block. Pearl's own wallets don't create these outputs yet; the team says a full post-quantum P2MR wallet is coming. Quantum computers don't break it the way they break elliptic-curve keys.

Post attachment
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 16

Pearl Supply check, week 29: 380.99M pearl-2:native emitted, 18.14% of the 2.1B cap. 19% is about 18M PEARL away. @PRLScan counted 1.10M PEARL issued in the last 24 hours.

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 16

Blockchain at Berkeley @CalBlockchain hosts the @prlnet research team this Tuesday, Nov 17, for a talk on proof of useful work, open to researchers, students and community members. pearl-2:native proof of work is matrix multiplication - the core operation in AI models - so a GPU can mine while it serves inference.

What if proof of work didn’t have to be wasteful? Join us this Tuesday (11/17) as we host the @prlnet research team for a conversation on Proof of Useful Work. Open to researchers, students, and community members! Food and drinks will be provided. RSVP: luma.com/mhctf8zr

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 16

Pearl pearl-2:native mining, week 29: Network hashrate 96.52 EH/s, up from 79.82 in week 28. @KryptexMining mined 1,505 of the last 7 days' 3,418 blocks, 44.0%, down from 48.0% in week 28. @perynai enters the top five with 206 blocks.

Post attachment
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 14

18% of Pearl’s 2.1B $PEARL cap has now been mined. The mark fell at block 142,733 on Nov 13, 17:14 UTC. The 17% mark came at block 133,179 on Oct 25, 19 days earlier. Pearl has no halvings: the block reward shrinks a little with every block, from about 2,225 PEARL at the 17% mark to about 2,172 at 18%.

Post attachment
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 14

The rules for @prlnet planned FP8 upgrade moved again. A change merged Nov 13 into the upgrade’s branch revises how the network judges a winning FP8 result. On B200 GPUs, the only hardware the code treats as downscaling, every row of the winning tile must keep a mean noise coverage of at least 1/25. An earlier, closed version of this change called the check “integer-attack security”: it targets a miner imitating FP8 math with cheaper integer arithmetic. The branch still has no pearl-2:native mainnet fork height.

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 13

pearl-2:native node software has a new release: pearld v1.5.7, tagged Nov 12. It adds an optional Prometheus metrics endpoint covering chain state, peer-to-peer traffic, RPC performance and the mempool, plus a health check. It’s off unless the operator sets --metricslisten. The docs note the endpoint has no authentication, so it should stay on localhost or behind a firewall. The release also hardens how nodes handle peers: message decoding, replies to peer requests, the signature cache, and peer timestamps, which no longer shift the node’s clock. The commit lists six security advisories as fixed. The release notes don’t say whether upgrading is required.

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 13

Pearl Research Labs has now officially announced PearlQuant, its KV cache compression method. It is @prlnet’s first post since Oct 21, and CEO Omri Weinstein calls it “the most efficient KV cache quantization on earth.” The results match the write-up this account covered on Oct 23. On GLM-5.3 it stores 3.2 bits per value against 9.1 for FP8, about 65% less, with no statistically significant quality difference on four benchmarks. At 128 concurrent sessions on one 8×B200 node, p90 time to first token is 7 s against FP8’s 144 s. Peak throughput is 24% above FP8’s. At light load, FP8 is about 8% faster, up from about 5%. The announcement is about serving and doesn’t mention mining. Mainnet mining still runs on 7-bit integer math, with the FP8 upgrade still to come.

Each inference session has its own KV cache, so a smaller cache means more concurrent users per GPU. Introducing PearlQuant, our new KV cache quantization technique: 65% less storage than FP8, better throughput and faster serving, same quality. GLM-5.3 on 8×B200, 128 concurrent sessions: ⚡ p90 TTFT: 144s → 7s 📈 24% higher peak throughput 🎯 Quality on par with FP8 pearlresearch.ai/research/bl…

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 12

Pearl Research Labs @prlnet has published what its team and shareholders hold in pearl-2:native. As of Nov 9: about 98M PEARL in total, which Pearl puts at about 4.6% of total supply. About 72M is held by Pearl Research on behalf of its team and shareholders, about 26M directly by shareholders. Pearl says all of it was mined or bought on the open market, that it has not sold PEARL to anyone, and that it and its team will lock at least 50% of their holdings for six months from Nov 9. For scale: the supply cap is 2.1B PEARL, and about 376.8M has been mined so far as of Nov 12.

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 12

New network hashrate high for pearl-2:native : 116.36 EH/s at block 142,085, Nov 12 08:11 UTC. That is prlscan’s 200-block rolling estimate, the same measure as the previous high of 106.04 EH/s on Nov 6.

Post attachment
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 11

pearl-2:native lowest published mining overhead, 3.9%, was measured on DeepSeek V3.2, a mixture-of-experts model, across 8 H200s. Mixture-of-experts certificates have been rejected on Pearl’s mainnet since block 91,630 (Jul 23), so that mode doesn’t mine today. For a dense model, the same Pearl post gives 5.08% (Llama 70B, 4 H200s), and @WeinsteinOmri put current integer mining at 10 to 15% on Sep 25. For the FP8 upgrade, still without a fork height, his newest figure is 1 to 5% end to end (Oct 8).

Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 11

The hash function under Pearl’s mining. The team wrote a GPU kernel that runs it about 370× faster than the BLAZE3 implementation, nearly 400× on large inputs, so hashing adds little to the AI work that mines $PEARL.

Post attachment
Documenting ¶Pearl avatar
Documenting ¶Pearl@DocumentingPRL·Nov 10

pearl-2:native inference platform lists the precision each model is served at: DeepSeek V4 Pro 0813, both DeepSeek V4.1 Flash models and Qwen3.8-27B run at FP8. GLM-5.3 and GLM-5.3-Flash run at FP4. Mining on Pearl's mainnet uses 7-bit integer math today, and the planned FP8 upgrade (still a draft, with no fork height) would move it to FP8. Mining on FP4 is listed as an open problem on Pearl's own Polymath page.