In May 2026 we announced an exclusive partnership with Together AI, and with it the first product where Pearl's economics reach an end customer directly: an inference endpoint priced more than 25% below the standard rate, offset by the PRL generated while the model runs.
This is the clearest existing demonstration of what Pearl is for. Not a theoretical claim about unit economics — a live endpoint with a lower price on the bill.
What was launched
| Model | Gemma-4-31B-it-pearl — an instruction-tuned, Pearl-certified variant |
| Where | together.ai/models/gemma-4-31b-it-pearl |
| Discount | Over 25% below the standard endpoint price |
| Announced | 15 May 2026 |
How the discount works
Ordinarily an inference provider's price has to cover GPU time, power, and margin. With Pearl, those same GPU cycles also produce PRL — so there's a second revenue stream from computation the provider was performing anyway.
Together AI passes that value back as a lower price per token. In their words, the discount is "offset by the future value of crypto emissions", and as the PRL spot price rises they expect to apply more of that value to reducing the endpoint price further.
Worth being precise about one thing: the discount is funded by the expected future value of emissions, not by revenue already banked. Together AI is taking a considered position on PRL's value in exchange for offering customers a lower price today. That's a real commercial judgement on their part, not an accounting trick — and it's why the arrangement can deepen as the asset matures.
As Omri Weinstein, our co-founder and CEO, put it: Pearl "increases the economic throughput of GPUs and decreases LLM price-per-token." The Together endpoint is that sentence made concrete.
Why this partnership specifically
Together AI runs production inference at scale for a large customer base, with serious research and inference-engineering teams behind it. Pearl's mining rides inside the model's forward pass, which means integration quality matters — a partner who understands their serving stack deeply is the difference between a real product and a demo.
It's also a useful proof of the open-weight thesis. Together AI's business is built on open-source models, which is exactly the population Pearl can certify.
What comes next
Together AI has said this is their first Pearl-integrated product and that they plan to expand the portfolio — including, eventually, ways for their customers to directly receive a portion of the currency emissions rather than only a discounted price.
That's a meaningful direction. It would move from "your inference is cheaper" to "your inference produces an asset you hold."
Does this affect PRL holders or miners?
Not mechanically. The partnership doesn't change the protocol, the emission schedule, or how mining works. What it does is put Pearl-powered compute into production at a real provider, which is the kind of adoption the network's long-term value depends on.
It also doesn't change your mining share directly — though as with any compute joining the network, more participants mean each existing miner's share of a block is smaller. See How mining on Pearl works.
I'm a developer — can I use the endpoint?
Yes. It's a standard Together AI endpoint — sign up with them and call it like any other model. You don't need a Pearl wallet, PRL, or any blockchain knowledge to use it. The Pearl part is invisible from the API's perspective; you just pay less.
I run an inference platform — can we do this?
That's the conversation we most want to have. See For inference providers: running Pearl-optimised vLLM, and note that the Pearl Inference Platform is currently in closed beta — join the waitlist.