Abakos Litepaper
Compute that pays for itself.
TL;DR. Abakos is a Proof-of-Useful-Work Layer-1 (a Bitcoin/btcd-style fork) where GPU miners get full block rewards only for compute that is actually paid for and used — enforced by an on-chain compute marketplace with stablecoin escrow. On top sits Abakos Chat, a "ChatGPT for everyone" payable with the token, turning mainstream demand into real, settled compute. Result: GPU hours cheaper than anywhere else, because the same watt-hour earns a mining subsidy and produces paid AI work (2-for-1) — economically enforced, not just promised.
The problem
- AI is bottlenecked by compute; it's expensive and centralized (AWS/Azure/GCP).
- Proof-of-Work blockchains burn enormous energy to produce only security.
- Proof-of-Useful-Work promised to merge the two — but the leading attempt (Pearl/PRL) only verifies that a matrix-multiplication was correct, not useful. Empirically: ~320,000 GPUs, 24 EH/s, ~112 MW producing zero useful AI — almost all random matrices.
The Abakos solution — 3 pillars
- Enforced usefulness (reward-split). Full rewards require work bound to a paid escrow job. Empty/random mining earns only 25%. Useful compute becomes the dominant economic strategy.
- On-chain compute marketplace + stablecoin escrow. Buyers lock payment; miners deliver output + ZK proof; settlement is atomic. The product Pearl never shipped.
- Standard-compatible & multi-vendor. BF16/FP8, drop-in for vLLM/SGLang, NVIDIA and AMD from day one.
Plus Abakos Chat: a simple, crypto-invisible AI app for everyday users → real paying demand flowing straight into the network.
How it works
A buyer orders compute (batch inference, embeddings, fine-tuning, or a chat session) and locks stablecoin in escrow. A miner picks up the job, runs the GEMM workload, and produces the AI output plus a PoUW/ZK proof binding the inputs to a registered model hash. Nodes verify the proof and the escrow; settlement pays the miner in stablecoin and grants the full block-reward share. Because the miner also earns ABK, the buyer pays below market — real 2-for-1 economics.
Token (ABK) & value accrual
- Utility: pay for compute / Abakos Chat (or stablecoin with ABK discount); mining rewards; optional staking; later governance.
- Value accrual: 2% marketplace fee (1% treasury / 1% burn) + burn of the empty-mining penalty → deflationary with real volume.
- Two separate flows: subsidy (token) vs. revenue (stablecoin) — kept strictly separate, technically and legally.
- Max supply: 2.1B ABK. Allocation: 70% mining, 10% treasury, 8% team, 7% liquidity & listing, 5% ecosystem (transparent vesting).
Note: Abakos does not issue its own stablecoin — it accepts existing ones (USDC/USDT).
Who buys compute
AI startups & indie devs (cheap inference), data teams (embeddings/labeling/synthetic data), everyday users via Abakos Chat, academic labs (subsidized GPU hours), and neoclouds reselling idle capacity. Fastest revenue: latency-tolerant batch jobs + Abakos Chat.
Differentiation
| Feature | Pearl | Gonka | io.net / Akash | Abakos |
|---|---|---|---|---|
| Useful work enforced | claimed | partial | n/a | ✓ economically |
| Compute marketplace | ✗ | partial | ✓ | ✓ + escrow |
| Consumer app (chat) | ✗ | ✗ | ✗ | ✓ |
| PoW price subsidy | ✓ (wasted) | ✓ | ✗ | ✓ toward real work |
Roadmap (mainnet-first, with a safety gate)
- P0–1: btcd fork, PoUW kernel, reward-split + model-hash commitment, public incentivized testnet (doubles as marketing).
- P2: marketplace MVP (batch), buyer SDK, Abakos Chat MVP.
- Audit (mandatory) of the new modules.
- P3: mainnet + token, DEX/CEX listing, FP8/BF16 + real-time chat path, neocloud partners.
- P4: vLLM/SGLang plugin, managed API, AMD/non-AI GEMM, Chat mobile + ABK pay.
← Join the waitlist · Join our Discord · Request the investor deck
Litepaper v0 — for information only; not financial advice and not an offer of securities. The ABK token structure is subject to legal review. Figures are illustrative.