Step 5 did WHAT? Testing Stepfun's model

I ran the same prompt on a folder of cute reference images through Step-5 Preview, DeepSeek Flash, and MiniMax M3.1 Flash. MiniMax went first, DeepSeek made its version, and then Step 5 took its turn. Step 5 already had an expression vocabulary, and it read my codebase and drew my sub-agents. Vision was the underrated part of the test, as you can see in my earlier five-model animation test.