SEAL gives you a library. Zama gives you a toolkit. H33 gives you a production-grade FHE API that processes 32 users in 939 microseconds — with integrated ZK proofs and post-quantum Dilithium signatures — in a single REST call. No PhD required.
The difference between a library and an infrastructure: one requires a PhD to deploy. The other requires one API call. Ten critical dimensions, four platforms, one clear winner.
| Feature | H33 | Zama (Concrete) | Microsoft SEAL | Duality |
|---|---|---|---|---|
| FHE Schemes | BFV, CKKS, BFV-32, FHE-IQ | TFHE only | BFV, CKKS, BGV | BFV, CKKS (OpenFHE fork) |
| Deployment Model | Managed REST API | Library (Rust / Python) | Library (C++) | Enterprise platform + consulting |
| ZK Proofs | Integrated ZK-STARKs (0.059µs) | No | No | No |
| PQ Signatures | ML-DSA (Dilithium) + ML-KEM (Kyber) | No | No | No |
| Biometric Matching | 939µs / 32-user batch | Build it yourself | Build it yourself | N/A |
| Published Throughput | 2.17M ops/sec (Graviton4) | Varies by workload | 10–50ms typical | Not published |
| Parameter Tuning | Automatic (FHE-IQ engine) | Manual — noise budgets, circuits | Manual — noise, encoding, moduli | Consulting-assisted |
| Open Source | Partial (benchmarks, specs) | Yes (BSD license) | Yes (MIT license) | No |
| Pricing | $0.033/op (volume discounts) | Free (you pay infra + engineers) | Free (you pay infra + engineers) | Enterprise licensing |
| Production Ready | Yes — live API, bare metal | Research / development | Research / prototyping | Enterprise pilots |
Open-source FHE libraries are brilliant research tools. They are not production infrastructure. Here is what ships with every H33 API call that the others simply do not have.
v10 production benchmarks from AWS Graviton4 (c8g.metal-48xl, 192 vCPUs, 96 workers). Every number is reproducible. Full benchmark data →
Encrypt data, compute on ciphertexts, and get a ZK-attested result — with post-quantum Dilithium signatures. While SEAL users are still configuring their noise budget, you are already in production.
# Encrypt biometric data and verify against stored template # FHE encryption happens server-side. ZK proof + Dilithium sig included. curl -X POST https://api.h33.ai/v1/fhe/verify \ -H "Authorization: Bearer h33_pk_..." \ -H "Content-Type: application/json" \ -d '{ "engine": "h33-128", "user_id": "user_abc123", "biometric": [0.23, -0.41, 0.87, ...], "options": { "zk_proof": true, "dilithium_attest": true } }' # Response (939µs FHE + 0.059µs ZKP + 291µs Dilithium): # { # "match": true, # "confidence": 0.97, # "zk_proof": "0x3a8f...", # "dilithium_sig": "ML-DSA-65:0xb4c1...", # "engine": "h33-128", # "latency_us": 1230 # }
const h33 = new H33Client({ apiKey: "h33_pk_..." }); // Encrypt + verify in one call. FHE-IQ auto-selects the engine. const result = await h33.fhe.verify({ userId: "user_abc123", biometric: embedding, // 128-dim vector, encrypted via BFV zkProof: true, attest: true, // Dilithium signature }); // result.match = true // result.zkProof = "0x3a8f..." (verifiable ZK-STARK) // result.dilithiumSig = "ML-DSA-65:0xb4c1..." (PQ audit trail) // result.latencyUs = 1230
Fair assessment. Real strengths. Honest trade-offs. We respect the work that Zama, Microsoft, and Duality have done for FHE. We just built something different.
The only FHE platform built for production API consumption at internet scale. Four FHE engines (H33-128, H33-256, H33-CKKS, H33-BFV32) with FHE-IQ auto-selection. Integrated ZK-STARK proofs and Dilithium/Kyber post-quantum signatures in a single REST call. 2.17M ops/sec on Graviton4 with 939µs biometric batches. Best for teams that refuse to spend a year building FHE infrastructure from scratch.
Open-source TFHE-rs (Rust) and Concrete-ML (Python) under BSD license. TFHE excels at boolean circuits, bit-level operations, and programmable bootstrapping. Concrete-ML makes encrypted ML inference approachable. Strongest choice for cryptography teams building novel FHE applications from scratch. Trade-off: single FHE scheme, no ZK proofs, no PQ signatures, no production deployment tooling, and you own all parameter tuning and infrastructure.
Mature MIT-licensed C++ library supporting BFV, CKKS, and BGV. The most widely cited FHE library in academic research. Well-documented with strong community. Trade-off: 10-50ms typical latency for operations H33 does in microseconds, requires deep understanding of parameter selection, noise management, and encoding strategies. No API, no deployment model, no ZK proofs, no PQ signatures. Best for researchers and prototyping.
Enterprise FHE platform built on an OpenFHE fork with professional services. Supports BFV and CKKS with a focus on regulated industries and data collaboration. Recently pivoted toward secure data sharing use cases. Trade-off: closed-source, enterprise-only pricing, no self-service API, consulting-heavy model means slower time-to-production. Best for large enterprises that need hands-on FHE guidance and have long procurement cycles.
Free tier includes 1,000 FHE operations per month. Four engines. ZK proofs. Dilithium signatures. One API call. No credit card required.