Mythic Network

Abstract

Mythic is a high-performance Layer 2 network built on the Solana Virtual Machine (SVM) using the Firedancer client. It introduces Proof of Useful AI Work (PoUAIW) — a novel consensus mechanism where validators earn rewards by performing real AI inference tasks rather than solving arbitrary computational puzzles. Every block produced on Mythic simultaneously advances the blockchain and contributes verifiable compute to a decentralized AI marketplace.

The network achieves peak throughput exceeding 1,000,000 transactions per second with 400ms block times by leveraging Firedancer's zero-copy networking architecture on dedicated hardware. All existing Solana programs deploy without modification on Mythic L2, providing immediate compatibility with the Solana ecosystem while adding native AI precompiles for inference, model verification, and logit validation at the execution layer.

Settlement occurs on Solana L1 through an optimistic rollup model with fraud proofs, ensuring that the full security of Solana's validator set serves as the final arbiter of Mythic L2 state. $MYTH, the native currency, powers gas fees, validator staking, AI compute payments, and on-chain governance with a fixed supply and deflationary burn mechanism.

1. Introduction

The intersection of artificial intelligence and blockchain technology has produced a wave of projects that bolt AI capabilities onto existing chains as afterthoughts. Most of these integrations suffer from fundamental architectural constraints: smart contract platforms were never designed to handle the data throughput, memory requirements, or latency demands of AI inference workloads.

Mythic takes a different approach. Rather than adding AI to an existing blockchain, we have built a blockchain that is AI-native from the ground up. The execution environment, consensus mechanism, validator economics, and system-level precompiles are all designed around the dual objective of processing transactions and serving AI compute — in the same pipeline, using the same hardware.

1.1 The Problem

Current decentralized compute networks face three interrelated challenges:

• —Verification gap: There is no trustless way to verify that a remote GPU actually ran the claimed inference correctly. Existing networks rely on reputation or economic bonds, neither of which provides cryptographic guarantees.
• —Latency mismatch: Blockchain settlement is too slow for interactive AI workloads. If every inference must wait for block finality, the user experience is orders of magnitude worse than centralized alternatives.
• —Economic misalignment: Traditional proof-of-work wastes energy on puzzles that serve no useful purpose. Proof-of-stake secures the chain but does not utilize validator hardware for productive compute during idle periods.

1.2 The Solution

Mythic addresses each challenge through architecture-level decisions:

• —PoUAIW consensus: Validators verify AI outputs by re-executing inference with deterministic sampling. The verification itself is useful work — checked outputs are cached and resold through the compute marketplace.
• —Firedancer-speed execution: Sub-second soft confirmations via the Firedancer sequencer allow AI applications to achieve latency comparable to centralized APIs while retaining on-chain settlement guarantees.
• —Unified economics: Validators earn $MYTH from both transaction fees and AI compute payments. Hardware required for block production (high-end GPUs, fast networking) is the same hardware required for inference — there is no idle capacity.

2. Architecture

Mythic L2 uses a modular architecture that separates transaction ordering, execution, AI verification, and L1 settlement into distinct layers. Each component can be scaled and upgraded independently while maintaining end-to-end security through Solana L1 as the data availability and settlement layer.

2.1 Execution Layer — Firedancer SVM

The execution layer is a modified Firedancer client running the Solana Virtual Machine. Firedancer's zero-copy networking, io_uring-based I/O, and NUMA-aware memory allocation deliver peak throughput exceeding 1M TPS on commodity server hardware. Mythic extends the standard Firedancer runtime with three additions:

• AI precompile programs (RunInference, VerifyLogits, RegisterModel) loaded as native system programs
• Modified tile layout with dedicated GPU inference tiles alongside standard networking and execution tiles
• Extended transaction metadata for AI job routing, model versioning, and compute attestation

2.2 Sequencer

The sequencer receives user transactions, orders them by priority fee, and creates batches for execution. It provides soft confirmations within 400ms — the user receives a provisional receipt before the batch is committed to an L2 block. The sequencer is currently centralized (operated by Mythic Labs) with a planned transition to decentralized sequencer rotation in Phase 4.

2.3 Data Availability

Transaction data is posted to Solana L1 in compressed form using erasure coding. This ensures that the complete L2 state can be reconstructed from L1 data alone, even if the Mythic sequencer goes offline. Full nodes can independently verify the L2 state by replaying L1 data.

2.4 Settlement on Solana L1

State roots are posted to the Mythic settlement contract on Solana L1 at regular intervals. The settlement contract validates state root commitments and manages the fraud proof challenge window. Any honest observer can submit a fraud proof within 7 days to challenge an invalid state transition, with successful challengers rewarded from the sequencer's bond.

3. Consensus — Proof of Useful AI Work

Proof of Useful AI Work (PoUAIW) is the core innovation of the Mythic network. It replaces the wasted computation of traditional proof-of-work with verifiable AI inference tasks, and augments the passive security of proof-of-stake with active utilization of validator hardware.

3.1 How It Works

3.2 Deterministic Inference

AI inference is not inherently deterministic — floating-point operations can produce different results on different hardware. Mythic solves this by requiring validators to use quantized model weights (INT8/INT4) and fixed random seeds for sampling. The combination of quantized weights and deterministic sampling ensures that independent executions produce identical logit distributions within a tolerance of ε = 10⁻⁵.

3.3 Model Registry

All models available for on-chain inference must be registered through the RegisterModel precompile. Registration includes a SHA-256 hash of the quantized model weights, an IPFS or Arweave URI for retrieval, and a minimum stake requirement for validators serving that model. Validators download and cache model weights locally; models are not stored on-chain.

4. Tokenomics

$MYTH is the native currency of the Mythic L2 blockchain. It is minted at genesis with a fixed supply of 1,000,000,000 tokens, 9 decimal places, and zero inflation. There are no team vesting schedules, no investor unlock cliffs, and no inflationary rewards. The supply can only decrease through burns.

4.1 Token Utility

4.2 Fee Distribution

All fees on Mythic L2 are denominated in native $MYTH. Each fee is split across three destinations:

4.3 Deflationary Mechanism

With 30–40% of all fees permanently burned, the circulating supply of $MYTH monotonically decreases. As network usage grows, the burn rate accelerates — creating reflexive scarcity. At projected mainnet throughput of 10,000 TPS average utilization, the annual burn rate exceeds 2% of total supply. There are no offsetting emissions; the supply curve is strictly decreasing.

5. Bridge & Settlement

The Mythic Bridge is a trust-minimized system for moving assets between Solana L1 and Mythic L2. It currently supports MYTH, SOL, and USDC, with additional Token-2022 compatible assets added through governance.

5.1 Deposits (L1 → L2)

Deposits are processed within seconds. Users send assets to the bridge contract on Solana L1. The Mythic sequencer monitors the bridge contract and credits the equivalent amount to the user's L2 address. Deposited assets are credited as $MYTH at the current exchange rate. Deposits benefit from Solana L1's ~400ms finality — funds are available on L2 within approximately 10 seconds.

5.2 Withdrawals (L2 → L1)

Withdrawals use an optimistic model with a 7-day challenge period. The user initiates a withdrawal on L2, which burns the L2 tokens and generates a withdrawal proof. After the challenge period expires without a successful fraud proof, the user claims their tokens on Solana L1. This delay ensures that any invalid state transition can be detected and challenged before funds are released.

5.3 Settlement Contract

The Mythic settlement contract (deployed on Solana L1) serves three functions: it records L2 state roots, manages the fraud proof challenge window, and releases bridged assets after the challenge period. The contract is secured by Solana L1's validator set — the most economically secure proof-of-stake network in production.

6. AI Precompiles

Mythic includes three native AI precompiles — system-level programs compiled into the execution layer that provide AI inference and verification capabilities to any Solana program running on the network.

7. Compute Marketplace

The Mythic Compute Marketplace is a decentralized exchange for GPU, CPU, and storage resources. Providers register their hardware, set pricing in $MYTH per compute unit, and receive payment as consumers use their resources. All settlement occurs on-chain through escrow contracts.

7.1 Provider Economics

Providers stake $MYTH as a quality bond and submit hardware attestations (GPU model, VRAM, CPU cores, storage capacity). The marketplace uses these attestations to match consumer requests with qualified providers. Providers are penalized (slashed) for failed job completions or incorrect outputs, ensuring reliability.

7.2 Pricing

By aggregating spare compute capacity from a global pool of providers, Mythic achieves prices 55–65% below centralized cloud providers for equivalent resources. Pricing is market-driven — providers compete on cost and reliability, and consumers benefit from transparent, on-chain price discovery.

8. Security Model

Mythic's security derives from three independent layers, each providing defense-in-depth against different attack vectors.

8.1 Solana L1 Settlement Security

The Mythic settlement contract on Solana L1 is the ultimate source of truth for L2 state. Any state transition posted to L1 can be challenged within 7 days by submitting a fraud proof. The challenge bond is set high enough to deter frivolous challenges while remaining accessible to honest observers. Successful challengers receive the sequencer's bond as a reward.

8.2 AI Verification Security

PoUAIW provides a second layer of security: validators independently verify AI inference results, catching any attempt to return fabricated outputs. Because verification uses the same model weights and deterministic sampling, an attacker would need to compromise a supermajority of validators simultaneously — an attack that is economically equivalent to a 67% stake attack on the entire network.

8.3 Economic Security

Validators and the sequencer stake $MYTH as collateral. Malicious behavior results in slashing: partial or total loss of staked tokens. The slashing penalty is calibrated to exceed the expected profit from any attack, ensuring that honest participation is always the dominant strategy. Slashed tokens are burned, further reducing supply and benefiting all remaining token holders.

9. Governance

Mythic governance is stake-weighted: one $MYTH staked equals one vote. Governance proposals can modify protocol parameters (fee rates, burn percentages, challenge periods), approve new bridged assets, adjust validator requirements, and schedule protocol upgrades. All governance actions are executed on-chain through the governance program.

The governance system will be activated in Phase 4 (Full Decentralization). Until then, protocol parameters are managed by Mythic Labs with community input through public RFCs and temperature checks.

10. Roadmap

11. Conclusion

Mythic represents a fundamental rethinking of what a blockchain can be. By making AI inference a first-class primitive of the execution environment and consensus mechanism, we eliminate the artificial boundary between “blockchain” and “compute network.” Every block produced on Mythic advances the chain, verifies AI outputs, and contributes capacity to a decentralized compute marketplace — simultaneously.

The combination of Firedancer's peak throughput (1M+ TPS), Solana L1's settlement security, a fixed-supply deflationary token model, and purpose-built AI precompiles creates a platform where AI applications can achieve performance comparable to centralized providers while inheriting the censorship resistance, transparency, and composability of a public blockchain.

Mythic is infrastructure for the next generation of AI — built on-chain, verified on-chain, and accessible to everyone.