Pearl (PRL) is a Layer-1 blockchain protocol developed by Pearl Research Labs, leveraging a Proof-of-Useful-Work (PoUW) consensus algorithm that replaces traditional hash-based mining with large-scale matrix multiplication, a core operation in AI training and inference. Its native token, PRL (¶PRL), is the network's currency. Its first major real-world validation came through a partnership with Together AI, an open-source AI inference provider ^{[1] [4]}.

​

Background and Development

Pearl was developed by Pearl Research Labs, led by co-founder and CEO Omri Weinstein, a complexity theorist with a PhD from Princeton University; he holds a research affiliation with Hebrew University and previously held roles at Nvidia and Vast Data ^[1]. The Pearl whitepaper, grounded in peer-reviewed cryptography research on matrix multiplication proofs, was published in April 2025 ^[2]. Mainnet launched on April 27, 2026 ^[2]. Pearl Research Labs' open-source code is publicly available on GitHub (pearl-research-labs/pearl) ^[8]. Pearl Research Labs claims a fair-launch model — no pre-mine, no founder token allocation, and no venture capital — similar in structure to Bitcoin's permissionless launch ^[1].

Technology — Proof-of-Useful-Work

Pearl replaces Bitcoin's SHA-256 hash puzzle with matrix multiplication (MatMul) as its proof-of-work function, using the same mathematical operation that underlies AI training and inference (the forward and backward passes of neural networks) ^{[5] [6]}. When GPUs run AI workloads, Pearl extracts valid proofs of those matrix multiplications as a side effect, allowing the same hardware to simultaneously secure the blockchain and perform AI computation.

The protocol combines MatMul commitments, BLAKE3 hashing, and zero-knowledge proofs to verify useful work without exposing private model weights or user data ^[2]. The current implementation focuses on exact integer MatMul; the white paper discusses a future path toward low-precision floating-point formats (BF16, FP8, FP4), but it does not include a fixed timeline for that upgrade ^[2]. Blocks are targeted every 194 seconds (approximately 3 minutes 14 seconds), which implies roughly 445 blocks per day under normal conditions. Mining initially required enterprise-grade GPUs such as Nvidia H100 and H200; consumer gaming GPUs were not supported in the initial implementation. ^[9]​

PRL Token

PRL (¶PRL) is the native cryptocurrency of the Pearl network. The token has a fixed maximum supply of 2,100,000,000 PRL (2.1 billion). Emissions follow a smooth polynomial curve defined as A(t) = t / (t + H), where H equals 650,226 blocks (approximately four years); under this schedule, approximately 50% of the total supply is scheduled to emit within the first four years ^{[1] [3]}. This contrasts with Bitcoin's abrupt four-year halving events and is intended to avoid sudden miner incentive cliffs while front-loading a large proportion of supply in early years ^[6].

As of approximately June 2026, block rewards were around 2,700 PRL per block, implying roughly 1.2 million PRL in daily issuance before transaction fees. ^[10] PRL reached an all-time high of approximately $1.65 on May 29, 2026, following the Together AI partnership announcement; price and per-GPU mining revenue subsequently declined as network difficulty rose ^{[7] [3]}. PRL is listed on ^[3].

Together AI Partnership (May 2026)

On May 15, 2026, Together AI, an open-source AI inference platform, announced an exclusive partnership with Pearl Research Labs ^[4]. The partnership launched Gemma-4-31B-it-Pearl, an instruction-tuned variant of the Gemma 4 31B model, as a serverless inference endpoint on Together AI's platform. The endpoint is priced more than 25% below Together AI's standard inference rates, with the discount offset by PRL emissions generated in parallel during inference computation ^[4]. Together AI had raised over 8 billion ^[1]. Omri Weinstein stated: "Pearl changes the unit economics of AI, by allowing every GPU cycle powering AI training and inference to simultaneously produce a native proof-of-work digital asset — the ¶PRL coin — at no additional cost" ^[4]. Together AI commented that as the spot price of PRL increases, more of the value generated by Pearl emissions could be used to further reduce inference prices ^[6].

The deployment represented the first production instance in which a commercial AI inference provider's GPU workloads simultaneously mined a proof-of-useful-work cryptocurrency ^[4].

Risks and Limitations

The HashRate Index analysis (June 2026) noted that the majority of current mining was inference that had not been purchased by end users, which indicates that current mining activity was largely AI-shaped proof-of-work rather than genuinely user-purchased AI computation. ^[10] Mining profitability declined rapidly after the Together AI announcement: an RTX 5090's estimated daily revenue approximately halved within weeks as network difficulty increased ^[7]. The smooth emission curve front-loads significant supply in the first four years, creating potential inflation pressure if miners sell rewards faster than markets absorb them ^[6]. PRL liquidity has been concentrated on smaller trading venues; the PRL ticker has historically been used by unrelated projects, creating potential for ticker confusion ^[5]. The broader founding team, legal entity, headquarters, and investor list were not fully disclosed in primary materials as of mid-2026. ^[10] The upgrade path to low-precision floating-point MatMul (BF16, FP8, FP4), which would broaden compatibility with standard AI workloads, had no disclosed timeline ^[6].