finally finished a full SlopCodeBench run against GLM 5.3, Sol 5.6, and Astra (Fable 5.1 results coming soon) This is different from all our previous research on SlopCodeBench (from @GOrlanski @ UW) in that we ran the full benchmark, every single scenario here. Previous runs did a small subset of the challenges. Asterisks: - this was run over ~1 week and had to be resumed a few times due to various provider outages - I still think a more scientific approach here would be to do what's common…