Developer tests Kimi K3 against Claude Opus in day of API runs

The tester ran ~20-30 API calls over one day, comparing Kimi K3 and Claude Opus on production-style workflows with identical prompts where possible. The post is framed as observations, not a benchmark or definitive comparison.