Research

Four papers, read for what they mean for PyrusLLM

Four independent groups, four disciplines — systems, cryptoeconomics, mechanism design, control theory — published between January 2025 and January 2026. None cite each other. Read together, they sketch the state of the art for decentralized LLM inference from every angle except one, and the missing angle is the one PyrusLLM is built around.

These are close reads, not abstracts: what each paper actually measured, where its numbers do and don't hold up, and — for each one — the specific design decisions it changes in PyrusLLM. Every page ends with concrete conclusions, not just a summary.

The four papers

🔌
PolyLink

PolyU Hong Kong + China Mobile HK, arXiv 2510.02395, Oct 2025. A decentralized LLM inference marketplace with a trustless quality-verification protocol (TIQE) and a three-role incentive model — the closest thing to PyrusLLM published as an academic system, with a real 20-device deployment.

⚡
DeServe

UC Berkeley (Dawn Song's group), arXiv 2501.14784, Jan 2025. Gives up on winning on latency and competes on offline batch throughput instead — 445 tok/s for a 70B model, pipelined across a high-latency WAN, with an optimistic verification scheme that costs one signature.

💰
AERIA

University of Exeter + HUST, arXiv 2503.04521, in review at IEEE TMC. An auction mechanism for edge inference pricing with proven truthfulness and envy-freeness, clearing in ~3 milliseconds — small enough to run inside a confidential enclave without discussion.

📈
DCBM

FLock.io + University of Oxford, arXiv 2601.09961, Jan 2026. A PID controller for token buyback-and-burn, positioned to stabilize the token that pays for all of the above — undercut by a simpler alternative the paper never considers.

The synthesis, in one paragraph

The three papers that have to match a request to a machine all leave that role centralized — two admit it in writing. AERIA is the most revealing: it doesn't leave the central coordinator in by accident — the auctioneer is the design, and the paper details exactly what it needs to know to clear the market: budget, deadline, accuracy requirement, and which model each user wants. That isn't an accidental leak. It's the algorithm's required input. It can't be encrypted, because the algorithm has to read it in the clear. It can only be moved somewhere nobody watching can see it — which is what a confidential-compute enclave is for, and is the thesis PyrusLLM is built on.

Where the four line up

PolyLinkDeServeAERIADCBM
Layerquality & incentivessystems & costprice discoverymonetary policy
Throughput / systemsweak (7 tok/s, sharded)strong (445 tok/s)n/a (CPU sim)n/a
Provider economicsinverted incentivereal cost tablereserve price + margin γn/a
Quality / reputationTIQE, measuredabsentdeclared σ, unverifiedn/a
Verificationcommittee, 30% permanent taxoptimistic, ~0% taxabsentabsent
Price discoveryundefined δfixed reference priceproven auctionstabilizes the unit
Privacynonenone (linkability is a stated feature)none — defines what the auctioneer must seen/a
Decentralized matchingno (cloud API server)no (admitted open problem)no (the auctioneer is the design)n/a
Real deployment20 devices, 4 citiesGCP east–westsimulation onlysimulation only
Public codeyesyesyesno
Applies to autoregressive LLMsyesyesno — single-forward-pass cost modeln/a

Sources

PaperarXivVenue / datePDF
PolyLink2510.02395v1cs.CR, 1 Oct 20252510.02395v1.pdf
DeServe2501.14784v1cs.DC, 4 Jan 20252501.14784v1.pdf
AERIA2503.04521v2in review, IEEE Trans. Mobile Computing2503.04521v2.pdf
DCBM2601.09961v1cs.GT, 15 Jan 20262601.09961v1.pdf

Read next

🎯
What to build first

The synthesis across all four: what to implement first, the three decisions that block everything else, and what not to build at all.

📜
Start with PolyLink

The closest published system to PyrusLLM — start here for the most direct comparisons.