Research11 min read
Unit-aware numerical reasoning: why LLMs quietly fail on dimensioned quantities
Large language models routinely drop, confuse, or invent units when reasoning over numerical problems. We show a reproducible failure mode, introduce a benchmark, and describe the inference-time approach Unitly uses to close the gap.
Analyticity Research
Research Team
Full paper and reproducible artifacts
The complete write-up — including methodology, benchmarks, ablation studies, and reproducibility notes — is being prepared for publication on this page. Enterprise partners and members of the research community can request early access while publication is in progress.
Permanent link: https://analyticitytech.com/research/unit-aware-numerical-reasoning