Unizen AI Publishes an Auditable ASI-Bench Result: Just Under 80 in Local Evaluation, With ...

Unizen AI published a verifiable ASI-Bench result scoring 79.86 out of 100 across 60 research tasks, outperforming all 18 state-of-the-art AI models tested in the benchmark, running entirely on an ordinary laptop CPU for about $0.20 with no model training required. The engine solves each task with analytic and first-principles models, providing a guidance-invariant approach and enabling auditability through hash-chained scorer receipts.

This Wire brief sits within Fusion42's coverage of AI Frontier Models.