Open-weight 31-35B models hit 87.9% task completion on agentic data prep.

A paper submitted July 23 benchmarks open-weight LLMs as coding agents across a consumer-grade deployment spectrum on 20 longitudinal data-preparation tasks producing 102 variables, reporting that current 31-35B models "almost saturated the benchmark" with average task completion up to 87.9%. The framework is open source and the argument is compliance: sensitive data never leaves the local environment. If data-residency rules block you from cloud agents, this is a concrete size target and a reusable eval harness. Single-source on a narrow domain, so read 87.9% as a ceiling for structured data-wrangling, not general coding.