Been seeing some chatter that the new mistral small 3.2 writes a lot like deepseek v3. This analysis of their slop profiles confirms. I think the network representation here makes a bit more sense than the phylo tree, given the complicated nature of model lineages.

New models benched: baidu/ERNIE-4.5-300B-A47B-PT and openrouter/cypher-alpha:free This mystery model is an odd one.

For a change of pace: some benchmarks Grok-4 didn't get SOTA on.

Let me spruik Judgemark real quick cuz I think it's neat: It works by measuring *separability* -- the ability to pin down writing ability in a blind test. The evaluated judge grades a set of models' writing outputs and we calc the error bar overlap. Less overlap: better judge.