Together AI launches a Pearl-powered inference endpoint with a stated 25%+ discount
A real partner integration and a direct test of the dual-output thesis. Lasting usage and economics are the next proof points.
What Together AI announced
Together AI said it launched an instruction-tuned Gemma model endpoint powered by Pearl’s proof-of-useful-work protocol, priced more than 25% below standard, with the discount offset by the expected future value of PRL rewards.
Together also said it eventually plans to expand its Pearl-powered lineup and to offer customers ways to obtain a portion of the rewards directly.
Why it matters
The partnership connects a real inference provider to Pearl’s central claim: one GPU cycle can answer a model request and contribute proof of work. That is stronger evidence than a miner multiplying random matrices with no customer.
What to track
- Endpoint availability and model coverage over time.
- Verified usage, latency, throughput and customer savings.
- How rewards are valued and sold in the discount calculation.
- Whether more providers reproduce the integration independently.
Sources for this story
Checked May 15, 2026
- Partner Together AI and Pearl Research Labs Team Up to Reduce the Cost of AI Inference Together AI (opens in a new tab)
- Partner Gemma-4-31B-it-Pearl model page (“not available on Together’s Serverless API”) Together AI (opens in a new tab)
- Primary Pearl network page Pearl Research Labs (opens in a new tab)
Software and protocol details change. Use LearnPearl to get oriented, then confirm anything involving money, keys or hardware at the official source.