Microsoft Magnetar Makes Zero-Knowledge Computing Up to 39 Times Faster
Microsoft's Magnetar zkVM proves Rust, Go and C programs up to 39 times faster than rivals, with 12 ms verification and security based on standard hashes.
Summary
Microsoft Research’s Magnetar is a general-purpose zero-knowledge virtual machine that proves RISC-V execution of normally compiled Rust, Go and C programs, then returns a compact certificate verifiable in about 12 milliseconds without rerunning code, trusting the operator or revealing private inputs. It addresses proof generation that historically cost millions of times more than native execution and circuit rewrites that took months. Applications include business policies, data pipelines, Ethereum Virtual Machine interpreters, AI-agent compliance, private regulated-finance processing, ledgers and blockchains.
On 15 laptop benchmarks spanning cryptographic hashing to token transfers in a small EVM interpreter, Magnetar proves 7.2 to 8.7 million processor instructions per second. It averaged 10 to 39 times the speed of the fastest comparable zkVMs on the same machine and exceeded two others by more than 300 times on token transfers. An optimized version reaches 9.5 to 12.8 million cycles per second, raising those speedups about 1.3 times.
Unlike systems using quantum-vulnerable elliptic curves, lattices, special-purpose hashes, trusted setup or unproven error-correcting-code conjectures, Magnetar relies only on SHA-256 or SHA-3 and binary-field arithmetic, aligning with NIST’s conservative hash-based post-quantum approach and Microsoft’s accelerated 2029 Quantum Safe Program goal. It culminates more than a decade of research: Spartan generalized proofs in 2019; Phalanx and Brakedown advanced speed and hash commitments in 2021; Nova, HyperNova, NeutronNova, Neo, SuperNeo, Spice, Lasso, Nebula, Twist and Shout, Jolt, BinarySpartan, Ligerito and ring switching supplied folding, memory, processor and binary-field techniques. Earlier in 2026, Vega proved government-credential facts in under 0.1 second.
Positives
- Magnetar verifies compact execution proofs in about 12 milliseconds without rerunning programs or exposing private inputs.
- Across 15 laptop benchmarks, Magnetar averaged 10 to 39 times the speed of the fastest comparable zkVMs.
- Optimized Magnetar reaches 9.5 to 12.8 million cycles per second, increasing reported speedups by about 1.3 times.
- Rust, Go and C programs can be compiled normally for RISC-V instead of being manually rewritten as cryptographic circuits.
- SHA-256 or SHA-3 security eliminates elliptic curves, lattices, trusted setup, special-purpose hashes and unproven coding conjectures.
- Self-verification lets Magnetar combine proof segments for computations of any size, including long-running ledgers and blockchains.
Risks & concerns
- Performance evidence covers 15 programs on a laptop, leaving production-scale results and hardware comparisons unspecified.
- No public release date, licensing model or deployment schedule for Magnetar is provided.
- Proof production historically required millions of times more computation than native execution, and Magnetar’s remaining overhead versus native code is unspecified.
- Existing elliptic-curve proof systems remain vulnerable to sufficiently powerful quantum computers, while newer alternatives may depend on less-tested assumptions.