PEARL's MoE Hard Fork Explained: What Changed and Why It Matters (2026)
In June 2026, PEARL completed its first coordinated hard fork — a network-wide upgrade that switched block certificates from the old "dense" model proofs to new Mixture-of-Experts (MoE) proofs, activating at mainnet block 71935. It's the kind of event that sounds deeply technical but actually says something important about where PEARL is going. This article breaks down what the MoE fork changed, why it matters for the useful-work thesis, and exactly what miners needed to do (spoiler: most needed to do nothing).
The 30-second answer
- What happened: At block 71935, PEARL switched from V1 (dense) ZK certificates to V2 (MoE) certificates. It was a coordinated hard fork — every node and pool had to upgrade beforehand.
- Why: To let the network mine Mixture-of-Experts models, the architecture that powers today's frontier AI (GPT-class models, Mixtral, DeepSeek, and most modern LLMs).
- For hashing miners: nothing to do. Old dense proofs still work before and after the fork. You only act if you want to mine the new MoE models.
- For the network: PEARL's "proof-of-work is real AI inference" thesis just got a major credibility upgrade — the work it produces is now current with what AI labs actually run.
First, what is a "Mixture of Experts" (MoE)?
To understand why this fork matters, you need 60 seconds on what MoE is.
A dense language model activates all of its parameters for every single token it generates. A 70-billion-parameter dense model does 70 billion parameters' worth of math per token. Powerful, but expensive.
A Mixture-of-Experts model is built from many smaller "expert" sub-networks plus a router. For each token, the router picks only a few experts to activate — say 2 out of 64. So a model can have hundreds of billions of total parameters but only activate a small fraction per token. You get the capability of a huge model at the compute cost of a small one.
This is not a niche trick. MoE is the dominant architecture at the frontier of AI in 2026 — it's how the largest, most capable models are built, because dense scaling hit a wall on cost. If a network's "useful work" is LLM inference, then supporting MoE isn't optional in the long run. It's the difference between serving the AI workloads of today versus the AI workloads of three years ago.
What the fork actually changed
PEARL's proof-of-work is unusual: the proof submitted to the chain is the matrix multiplication done during LLM inference (the NoisyGEMM kernel on NVIDIA tensor cores). If you're new to that idea, read our explainer on Proof-of-Useful-Work first — it's the foundation for everything below.
Because the proof is tied to the model being run, supporting a new model architecture (MoE) required changing the cryptographic certificate that proves the work was done. That's a consensus-level change — hence a hard fork. Specifically:
- Before block 71935: blocks carried a V1 (dense) certificate.
- At and after block 71935: blocks carry a V2 (MoE) certificate.
- The V2 prover is backward-compatible — it accepts both old dense proofs and new MoE proofs. So miners running dense models kept working through the fork without interruption.
The upgrade touched two pieces of infrastructure: the node software (upgraded to v1.1.0) and the ZK proving code that pools use to turn a miner's raw proof share into the certificate that goes in the block. Both had to be deployed before the fork height, or blocks built after the fork would be rejected.
What miners needed to do
This is the part that surprised people — for most miners, the answer was nothing.
| You are… | What you had to do |
|---|---|
| A hashing miner on a pool (renting GPUs, pointing them at AlphaPool, LuckyPool, PearlHash, etc.) | Nothing. Your pool handled the node + ZK proving upgrade. Your dense miners kept working before and after the fork. |
| A pool operator | Upgrade the node to v1.1.0 and the ZK proving code before block 71935. This was the real work — and it fell on the pools, not individual miners. |
| Running your own full node (e.g. for an explorer or service) | Upgrade the node to v1.1.0 before the fork, or it stops following the chain at block 71935. |
| Want to mine the new MoE models | Optional, and only after the fork: upgrade your miner software to produce MoE proofs. If you keep mining dense, nothing changes. |
The smooth-rollout design is worth noting: a hard fork that requires zero action from the people running the actual hardware is a sign of a mature upgrade process. The disruption was concentrated on a handful of pool operators who coordinated with the core team, not spread across thousands of miners.
Why this matters more than it looks
1. The "useful work" stays genuinely useful
The strongest argument for PEARL is that its energy doesn't just secure a chain — it produces real LLM inference (see our energy comparison with Bitcoin). But that argument only holds if the inference being produced is current. A network locked to dense-only models would slowly become a museum piece — burning power to run an architecture the AI industry has moved past. The MoE fork keeps PEARL's output aligned with what frontier labs actually deploy. That's the difference between "useful work" being a real claim versus a historical footnote.
2. It proves the chain can evolve
This was PEARL's first coordinated hard fork. Executing one cleanly — with advance notice, a backward-compatible prover, a published upgrade guide, and coordination with every major pool — demonstrates the protocol can ship consensus-level changes without chaos. For a network whose entire value proposition is tracking the moving target of AI compute, the ability to upgrade smoothly is not a nice-to-have. It's existential. The first fork landing cleanly de-risks every future one.
3. It widens the moat
Supporting MoE proofs is hard. It requires the ZK proving system to certify a fundamentally more complex computation (variable expert routing) while keeping verification cheap. Most "AI x crypto" projects can't do this because their token is just a payment layer bolted onto an API — there's no proof system to upgrade (see what mining PEARL actually produces). PEARL shipping an MoE-capable proof system is a capability very few networks could replicate.
The timeline
| When | Event |
|---|---|
| Early June 2026 | Core team announces the coordinated MoE hard fork; publishes the upgrade guide |
| ~24h before fork | v1.1.0 node + ZK proving code released; pools given ~24h to upgrade |
| Block 71935 (mid-June 2026) | Fork activates. Blocks switch to V2 (MoE) certificates |
| After the fork | Miners can optionally deploy MoE-capable miners; dense mining continues unchanged |
The network passed through the fork without disruption — block production continued, dense miners kept earning, and the chain is now MoE-ready. You can watch live network activity any time on the Lord Of Pearls explorer and the live pool comparison.
Frequently asked questions
What does MoE stand for in the PEARL hard fork?
Mixture of Experts — a neural-network architecture where each token is routed to a small subset of specialized "expert" sub-networks instead of activating the whole model. It's the dominant architecture for frontier LLMs in 2026. The fork lets PEARL mine (and therefore serve inference for) MoE models, not just dense ones.
Did I need to do anything as a PEARL miner?
If you mine on a pool (the vast majority of miners), no — the pool handled the upgrade and your dense miners kept working. You only need to act if you want to mine the new MoE models, and only after the fork. If you run your own full node, you needed to upgrade it to v1.1.0 before block 71935.
What happens to old dense miners after the fork?
They keep working. The V2 (MoE) prover is backward-compatible and accepts dense proofs. Dense and MoE shares are both valid after the fork. Nothing was deprecated for hashing miners.
What was the fork height?
Mainnet block 71935. Nodes and pools were told not to hardcode the height but to read the required certificate version from the block template (the node exposes a requiredcertversion field), so the switch happened automatically at the right block.
Is mining MoE more profitable than mining dense?
It depends on demand for MoE inference versus dense inference and the relative difficulty. As of the fork it's too early to say — both are valid and the market for each will set the economics. For most miners, the practical answer is to keep mining whatever your pool supports and let the pool optimize. See the pool comparison for which pools support what.
Does the MoE fork change PEARL's energy profile?
Not materially. MoE models activate fewer parameters per token, so per-token inference is cheaper in compute — but mining difficulty adjusts to target block time regardless. The headline takeaway from the energy comparison holds: PEARL's energy still produces real inference as a by-product of securing the chain.
Where do I buy or sell PRL?
PEARL still has no CEX listing (June 2026) — all trading happens on OTC desks. The lowest-fee venue is Lord Of Pearls OTC at flat 1.8% per trade. See the full exchanges comparison for all three PRL/USDC venues side by side.
Bottom line
The MoE hard fork looks like a technical footnote and is actually a milestone. It keeps PEARL's useful-work output current with frontier AI, it proved the chain can ship coordinated consensus changes cleanly, and it widened a moat that's already hard to cross. For miners it was a non-event by design — dense mining never skipped a beat. For the thesis, it was a quiet but significant step: a proof-of-useful-work chain is only as compelling as the usefulness of the work, and PEARL just made sure that work stays on the frontier.