I Tested GLM-5.3-Flash and Qwen3.8-Flash on 24 Real Tasks

I test-ran both of this week's open-weight flash models against 24 small, real workloads from an...