Pearl MoE Hard Fork Goes Live, All Nodes Must Upgrade

Steady attention without excessive speculation.

Pearl Research Labs announced the upgrade on X at roughly 06:51 UTC. According to the team, the fork is already active. Node operators must update now or get left on an incompatible chain.

What the MoE Hard Fork Changes

The core change is the certificate format. Pearl is a Proof-of-Useful-Work chain. A certificate is the verifiable proof that a miner did real compute.

Until now, miners produced v1 certificates. These covered exact integer matrix multiplication, or MatMul, run on perturbed inputs. The official announcement says v2 certificates instead support Mixture-of-Experts, or MoE, workloads.

MoE is a common AI model design. Instead of running every input through one giant network, the model routes each input to specialized “expert” sub-networks. As a result, only part of the model activates at a time.

According to Pearl, v2 will “more accurately verify and reward useful AI computation.” The goal is simple. Tie mining rewards to genuine AI work, not arbitrary math.

Why Nodes Must Upgrade Now

The Pearl MoE hard fork rewrites the rules for valid blocks. Because of that, old software cannot validate the new v2 certificates.

The upgrade call asks operators to run pearld v1.1.0 and wallet v2.0.0. The wallet release bundles the fork logic through pull request #173 on the project’s GitHub releases page.

Nodes that skip the update will follow the old ruleset. Therefore they will split off from the canonical chain. For miners, that means lost rewards until they upgrade.

How Pearl’s Proof-of-Useful-Work Functions

Pearl is a Bitcoin fork, but it does not use SHA-256 hashing. Instead, miners run large matrix multiplications on Nvidia GPUs. These are the same operations that power AI training and inference.

The design traces back to a 2025 paper, “Proofs of Useful Work from Arbitrary Matrix Multiplication.” Pearl CEO Omri Weinstein co-authored it. At mainnet launch on April 27, the team shipped a vLLM miner plugin so GPU owners could join.

Each miner commits to input matrices, adds controlled low-rank noise, then computes the noised MatMul. Next the miner generates a zkSNARK proof using Plonky2. Hash conditions must also meet network difficulty.

Verifiers check the proof without rerunning the full computation. So the network confirms the work happened and then pays the miner in PRL. Pearl calls this a “2-for-1” model, where security and useful compute come together.

Why MoE Workloads Matter for Rewards

The shift to MoE is about reward accuracy. Under generic MatMul, a miner could pick convenient matrices. Those matrices are easy to compute but not tied to any real model.

MoE-structured workloads are harder to fake. The work mirrors actual sparse model inference. So Pearl argues v2 certificates can reward compute that maps to productive AI tasks.

Still, the exact technical spec for v2 is not yet public. The whitepaper details the v1 MatMul design. Detailed v2 documentation has not surfaced, so the announcement remains the main source.

PRL Price and Market Reaction

PRL traded near $0.509 around the time of the fork, according to CoinGecko data. The 24-hour change sat near flat. The 7-day change showed a drop of roughly 33%.

Daily volume was modest at about $591,000, which points to thin liquidity. PRL carries a max supply of 2.1 billion coins. Trackers do not yet report full circulating supply. Third-party sites put network hashrate near 28 EH/s in early June.

So far no tier-1 outlet such as CoinDesk or The Block has covered the fork. The event is only hours old, so broader market reaction may still develop. None of this is financial advice.

Skeptics Question the Useful Work Claim

Pearl has its doubters. A recent arXiv paper studied Pearl’s cuPOW protocol. A Hacker News thread joined in. Both question how much real AI inference the network actually performs.

Critics argue that miners may still favor convenient matrices over genuine workloads. Some also note that Together AI, Pearl’s inference partner since May 15, reportedly subsidizes its own inference separately.

Pearl frames v2 as a direct answer to that critique. Whether MoE certificates close the gap between claimed and actual useful work remains an open question for now.

What Comes Next for the Pearl Network

The Pearl MoE hard fork is the network’s first major upgrade since mainnet. The immediate task is adoption. Pearl needs miners and node operators to move to v1.1.0 quickly. Clean upgrades will decide how smooth the transition feels.

Several details are still unconfirmed. Pearl has not published the exact fork activation block height. Post-fork node counts and any v2 security audit also remain unclear.

Miners can track new blocks on the Pearl block explorer as v2 certificates roll in. Pearl built its pitch on the promise of useful AI compute. This fork is the first real test of that thesis.