STARKs vs SNARKs em 2026: comparativo técnico de zero-knowledge proof systems
Comparativo técnico entre STARK e SNARK proof systems — assumptions criptográficas, proof sizes, prover times, e quais sistemas são usados por quais ZK-rollups em 2026.
29/03/2026 · 7 min de leitura · ZK · Tecnologia · DeepDive · Cryptography
Conceito básico ZK
Zero-Knowledge Proof (ZKP) permite que uma parte (prover) prove que sabe valor X que satisfaz certain condition, sem revealing X. Em blockchain, isso permite proving that a batch de transactions é valid sem replaying todas as transactions on-chain — base de ZK-rollups.
Two main families em production cripto: SNARK (Succinct Non-interactive ARgument of Knowledge) e STARK (Scalable Transparent ARgument of Knowledge). Cada family tem several specific protocols (Groth16, Plonk, FRI, etc).
SNARKs: pros e cons
Pros: Smallest proof sizes (200-500 bytes typical Groth16; 1-2 KB Plonk). Fast verification (sub-millisecond). Mature tooling (Circom, snarkJS, gnark).
Cons: Trusted setup ceremony required for most SNARK constructions (Powers of Tau era done em 2023 for Ethereum-related; specific circuit setups required per circuit). Trusted setup is single point of failure conceptual — if compromised, all proofs sob esse setup são fakeable.
Used by: Aztec Network (Plonk), Polygon zkEVM (Plonky2 — hybrid), zkSync Era (custom Plonk variant), Loopring (Groth16).
STARKs: pros e cons
Pros: No trusted setup needed (uses public randomness — sometimes called 'transparent'). Post-quantum secure (relies on hash functions, not elliptic curve assumptions vulnerable a quantum). Larger circuits scalable.
Cons: Larger proof sizes (40-200 KB typical). Slightly slower verification (10-50ms). Less mature tooling than SNARK ecosystem.
Used by: Starknet (primary deployment), StarkEx (Immutable X, dYdX V3, Sorare). Polygon Miden (in development). Risc Zero zkVM partially uses STARK techniques.
Proof systems específicos
Groth16 (SNARK, 2016): smallest proof (200-300 bytes), fastest verification. Trusted setup per circuit. Used in early ZK applications (Zcash, Tornado Cash).
Plonk (SNARK, 2019): universal trusted setup (one setup serves many circuits). Slightly larger proofs (1-2 KB), still fast verification.
FRI (STARK, 2018): no trusted setup. Used by Starknet em Stone prover.
Plonky2 (Hybrid, 2022): combines Plonk efficient verification with FRI's no-setup. Used by Polygon zkEVM.
Halo2 (SNARK, 2020): recursive proofs, no trusted setup needed (custom KZG). Used by Zcash Orchard.
Performance benchmarks em 2026
Numbers below são approximations baseadas em research papers e team disclosures, vary significantly por circuit complexity:
Groth16: prover time ~30-60s per million constraints, proof size 200-300 bytes, verification <10ms.
Plonk: prover ~60-120s per million constraints, proof size 1-2 KB, verification <20ms.
STARK (FRI): prover ~10-30s per million constraints (faster!), proof size 40-100 KB, verification ~20-50ms.
Plonky2: prover ~15-40s per million constraints, proof size 5-15 KB, verification ~15-30ms.
Trend: prover times caindo rapidamente com GPU acceleration. ZK Hardware Acceleration via specialized chips (Ulvetanna, Cysic) reducing prover time orders of magnitude.
ZK-rollups em produção
Starknet: STARK-based (FRI). TPS sustained ~100, peak ~500. Volume ~150k transactions diários em maio 2026. Fees ~$0,01-0,05 per tx.
zkSync Era: SNARK-based (custom Plonk). TPS sustained ~50, peak ~200. Volume ~300k transactions diários. Fees ~$0,02-0,10.
Polygon zkEVM: SNARK-based (Plonky2). TPS sustained ~30. Smaller than competitors em volume.
Linea (Consensys): SNARK-based (PLONK). Mainnet 2023. Growing user base.
Scroll: SNARK-based (Halo2/Plonk-derivative). EVM-equivalent (highest level compatibility).
Trade-offs e o futuro
Para developer: choose proof system based em use case. Need fastest verification? Groth16. Need post-quantum security ou no trusted setup? STARK. Need balance e ecosystem support? Plonk.
Para user: choose ZK-rollup based em ecosystem, fees, application support — proof system underneath é abstraction.
Future direction: hybrid systems combining best of each (Plonky2 é exemplo). Hardware acceleration making prover times negligible. ZK becoming general-purpose computing layer.
Aviso YMYL: ZK technology is rapidly evolving. Specific implementations may have unforeseen bugs. ZK-rollups têm shorter security track record than optimistic rollups. Diversify between L2 categories para significant holdings.