01 · 20%
Reasoning
Multi-step deduction, causal inference, and structured decomposition of ambiguous problems.
Highest weightQuan Bench profiles leading AI models across six dimensions of real business intelligence. Not synthetic benchmarks - applied judgment. We publish our scores openly.
Six axes. Weighted by business impact. Evaluated across 200 standardized prompts per dimension, re-run on every major model release.
Multi-step deduction, causal inference, and structured decomposition of ambiguous problems.
Highest weightFactual correctness and hallucination resistance across a curated set of verifiable knowledge tasks.
Highest weightCoherence over long-form, multi-constraint prompts and extended conversation chains.
High weightConsistency across repeated identical prompts and resistance to adversarial edge cases.
High weightPrecision over volume - delivering concise, targeted responses without unnecessary verbosity.
Medium weightNovel framing and original output generation under open-ended, unconstrained briefs.
Base weight