MiniMax said on its official blog on November 20 that its research team used an internal-only research model, MiniMax's Internal Model (MIM), for automated cryptanalysis. On CryptanalysisBench Tier 2 it produced the first practical breaks of several NIST competition candidates that had no known practical attacks, found new attacks on peer-reviewed or award-winning designs such as COLM, SBC-VOLE and the NSA's ARADI, and solved 55 previously unsolved TII McEliece Challenge instances. MiniMax stresses the work covers academic designs, past candidates and public challenge parameters and does not affect deployed standards. Technical reports and proof-of-concept code are on GitHub; the model itself is not public, and researchers can apply for test access.

Key Takeaways

  • ✓NIST candidates: collisions on SHA-3 candidate Waterfall-256 across all output lengths in 0.045s on one core; full-round Blender-256 collisions in 0.36-2.3s
  • ✓Post-quantum: LOCKER solver matrix shrunk from 42 PB to 58 MB with the shared secret recovered in 2s; ROLLO-I-128 equivalent secret key from the public key in about 6s
  • ✓Implementation bugs: more than a dozen unreported flaws in reference code, e.g. DRS mistyped && as &, enabling signature forgery with zero signing queries
  • ✓Peer-reviewed designs: COLM forgery from two encryption queries; SBC-VOLE full key from one signature in 0.06s; ARADI practical key recovery pushed from 5 to 8 rounds
  • ✓Caveats: results are self-reported and the model stays internal; MiniMax says deployed standards are unaffected and scheme authors were notified in advance
MiniMax Reveals Automated Cryptanalysis by Its Internal MIM Model: First Practical Breaks of Several NIST Candidates, Waterfall-256 Collisions in 0.045s and Up to 43,400x Faster McEliece Solving
🖼️Official Media
Click to view high-res
🧭

Turn your technical choice into a development budget

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

🔬

In-Depth Technical Analysis

Background

Cryptanalysis forces a model to turn math into attack code that either works or fails, which makes it a hallucination-proof test of long-horizon reasoning. MiniMax published results from its internal research model MIM on its official blog on November 20, and accounts on X such as __jingling are discussing them.

How it works

The blog does not disclose MIM's architecture or size. The team used the CryptanalysisBench setup (NeurIPS 2026; ETH Zurich, Anthropic and collaborators) on Tier 2: 135 publicly reviewed schemes with no known practical attacks plus 7 production-strength challenge schemes. A separate automated framework targeted papers from top venues and competition winners.

Numbers that matter

Waterfall-256 collisions in 0.045s; LOCKER shared secret in 2s; COLM forgery from two encryption queries; ARADI practical key recovery extended to 8 rounds. On the TII McEliece Challenges it recovered keys for 94 instances, 55 previously unsolved, solving instances with prior records in 1.80-5.36s on one 64-core server, 6,180x to 43,400x faster than the CRYPTO 2026 record (9.20 hours). For HAWK it cut bit-security estimates by another 9.3-23.8 bits beyond Anthropic's analysis. All figures are self-reported with no independent reproduction yet.

What it means for developers

Reports and PoC code are at GitHub MiniMax-AI/Research for anyone to reproduce. MIM is not public; researchers can request test access via [email protected]. MiniMax says deployed standards are unaffected, but teams shipping custom or niche cryptography should consider AI-assisted review of their reference implementations.

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

Up NextX Discussion: Ex-OpenAI Researcher Will DePue Publishes a Checkpoint Forensics Report Claiming Six Open-Weight Models from DeepSeek, Moonshot, MiniMax and Ant Group Carry NVIDIA Blackwell Initialization Fingerprints; the Report Says It Does Not Prove Export-Control Violations
Read Now→