Simon Willison Releases smevals: A Lightweight Eval Suite for LLMs

Simon Willison · rss · 2026-08-01

Simon Willison, working with Prime Radiant lab, has released smevals, a lightweight eval suite designed to evaluate the capabilities of different models, prompts, and harnesses.

The tool is built for AI coding agents: developers can simply instruct their agent to run uvx smevals docs to learn the tool and build an eval suite. It decouples runs from grading operations, uses YAML files for checks, and includes a local web server or static HTML builder for exploring results.

Original post →

More from coding & agent

coding & agent channel →