benchmarksfactbullishA new tool called 'smevals' enables running small eval suites against models, harnesses, and promptsSimon Willison01 Aug 2026https://x.com/simonw/status/2083310510729216039