As reported on November 19, 2026, Pearl had activated an FP8 upgrade on its testnet the previous day. The Bitcoin code fork proposes a form of “Duplex Work”: an AI inference provider could submit proof of computation performed for customers and receive PEARL block rewards. The test is whether the network can distinguish that useful work from computation performed solely to earn tokens.
What FP8 changes
Bankless analyst David Christopher’s account describes a shift from an earlier mining mechanism that supported INT7 to one that accepts FP8. Under the earlier mechanism, providers using other precisions would have needed to convert their models to INT7 to participate. The report says more than 80% of model-inference traffic on OpenRouter was running in FP8, suggesting the change could reduce an integration barrier for some providers. It does not supply a measurement date or underlying dataset for that percentage.
The trade-off is hardware access: according to the account, INT7 can run on a wider range of older graphics cards, while FP8 generally requires newer GPU architectures. More importantly, FP8 compatibility does not establish that a submitted computation served a real customer. The report says Pearl did not require Duplex Work at that stage, leaving miners able to earn rewards without performing genuine AI inference. It also says stronger on-chain verification would be needed to address that problem.

What remains unproven
The report proposes three ways to assess the model over the following months: whether established inference providers publicly confirm they mine PEARL and pass token subsidies through as API discounts; whether mining power becomes less concentrated; and what share of blocks corresponds to inference requested by real users. It identifies the last measure as especially difficult for outsiders to observe. These are proposed tests, not reported results.
PEARL had risen over the preceding week despite a weaker broader crypto market, according to the report, which also notes Coinbase and Bybit listings. No price series or figures are provided to quantify the move or isolate its cause.
Crypto researcher hitesh.eth offers a more skeptical interpretation. He argues that the investment narrative depends on Bitcoin miners moving into AI compute, that token inflation could make rewards attractive only temporarily, and that coordinated influencer promotion appears to be driving attention. Those are his assessments, not established findings about miner migration, promotional coordination or product adoption. For now, the reported testnet upgrade demonstrates a change in eligible computation—not that Pearl has verified customer inference at scale or built a durable market for it.

