Pearl Rewards Miners for AI Math Without Proving Customer Work

Pearl miners earn the Pearl token (PRL) for proving matrix multiplication, but the network cannot verify that the calculations serve customer AI jobs, a distinction central to its useful-work model.

Pearl is a Bitcoin (BTC)-derived network with an unspent transaction output (UTXO) design. Matrix multiplication combines rows and columns of numbers and is used in AI workloads. Miners choose their own matrices, and nodes can check the calculation without seeing its inputs. The protocol does not require those inputs to come from a customer’s model or inference job.

Abhinaba Basu’s June 2026 preprint, which is under review, analyzed 8,012 AlphaPool workers over May 30–31. It estimated the pool represented about 21% of Pearl’s network hashrate during that period and found no identifiable inference code in its dominant mining software. The preprint also described a random-matrix miner using NVIDIA and AMD hardware that earned 44 pool-accepted shares. The findings cover one pool over two days and do not establish how every Pearl miner operated.

Pearl’s code caps issuance at 2.1 billion PRL and lowers block rewards gradually, without fixed halving events. The target block interval is 194 seconds. At block 128,999, timestamped 12:54 a.m. ET (04:54 UTC) Oct. 17, the chain had issued 347.65 million PRL, or 16.6% of the cap. The block reward was 2,248.84 PRL, equivalent to about one million new PRL a day at the target interval. The schedule issues tokens regardless of whether a calculation can be tied to customer work.

Early blocks arrived much faster than the target pace. Block 40,000 was recorded at 2:31 a.m. ET (06:31 UTC) May 3, fewer than six days after the network launched on Apr. 27. Applying Pearl’s emission formula gives about 121.7 million PRL issued by that height; at the target interval, 40,000 blocks would take roughly 90 days. Pearl set initial mining difficulty low in part to provide the core team with resources for research and development.

Together AI announced a Pearl-related endpoint for Gemma-4-31B-it-pearl on May 15, describing it as its first product to integrate Pearl. The company said the endpoint was discounted by more than 25%, with the reduction offset by the future value of crypto emissions. The announcement did not quantify how much Pearl mining serves paid inference requests. Whether customer AI workloads become a meaningful part of mining remains central to Pearl’s useful-work model.