Benchmarks Stack Ranking
APIs Pricing Docs White Paper Token Blog About Security Demo
Log In Get API Key
AWS Graviton4 · ARM Neoverse V2 · c8g.metal-48xl (96 cores) · 2,172,518 auth/sec sustained · SIMD batching: 32 users/ciphertext · ~42µs per auth
Live Cryptographic Verification

Verify It's Real

Run live post-quantum cryptographic operations against the production backend. Real FHE encryptions. Real STARK proofs. Real Dilithium signatures. Every result is independently verifiable.

Get Free API Key → Schedule a Demo
H33-128 BFV · NIST L1

FHE encrypt inner product decrypt. 128-bit lattice security with BFV scheme.

Idle
Latency
Signature
Proof Hash
Timestamp
H33-256 Level 5 · 256-bit

Same pipeline at 256-bit security. CRYSTALS-Dilithium Level 5 signatures.

Idle
Latency
Signature
Proof Hash
Timestamp
Full Pipeline FHE + ZKP + PQC

FHE encrypt + STARK proof + Dilithium attestation. Complete 7-layer post-quantum pipeline.

Idle
Total Latency
FHE Latency
ZKP Latency
Signature
Proof Hash
Timestamp
5 requests per minute · Production Graviton4 backend

What You're Testing On

Every cryptographic operation above executes on AWS Graviton4 (ARM Neoverse V2 microarchitecture) — the same production infrastructure that sustains 2,172,518 authentications per second. This is not a simulation. The API at api.h33.ai runs our full post-quantum pipeline in Rust on bare-metal ARM silicon.

~42µs
Per Authentication
2.17M
Auth/sec Sustained
96
Dedicated Cores
32
Users per Ciphertext

Why Graviton4 + Batching Makes This Faster

The metal-48xl instance provides 96 dedicated physical cores (not shared vCPUs) with the full Neoverse V2 instruction set, including NEON SIMD. Our Rust engine exploits this at every layer:

Montgomery NTT — Number Theoretic Transform with twiddle factors in Montgomery form. Zero division in the hot path. All polynomial multiplications stay in NTT domain until final INTT.
NEON Galois Ops — Branchless permutation and vectorized key-switching using ARM NEON intrinsics. Galois rotation in the FHE ciphertext space runs at near-hardware speed.
Harvey Lazy Reduction — Butterfly values held in [0, 2q) between NTT stages. Halves the modular reduction work per stage, compounding across 12 NTT levels.
SIMD Batching — 4,096 polynomial slots ÷ 128 biometric dimensions = 32 users per ciphertext. One FHE multiply handles an entire batch. Amortized per-auth cost drops from ~1.36ms to ~42µs.
Batch Dilithium Attestation — One CRYSTALS-Dilithium sign+verify per 32-user batch instead of individual signatures. 31× reduction in post-quantum signature overhead.
NTT-Domain Persistence — Enrolled templates stored in NTT form. Public key pk0 pre-NTT'd at keygen. Eliminates redundant forward transforms in the verify path.

The result: a full 7-layer post-quantum pipeline (FHE encrypt → BFV inner product → STARK lookup proof → Dilithium attestation → decision) completes in ~1.36ms single-thread. With 96-worker parallelism and 32-user batching, the metal instance sustains 2,172,518 auth/sec (±0.3% run-to-run variance). That's ~42µs per authentication — 220× faster than a human blink.