One month coding with GLM 5.3 Flash

The author spent September using primarily GLM 5.3 Flash for coding, consuming 2 billion tokens total, but encountered significant challenges that forced deviations from the plan. Major setbacks included inefficient model selection for prototyping (costing $150 and 450M tokens), infrastructure availability issues with popular inference providers, and the need for continuous experimentation across multiple models for benchmarking purposes. Despite the budget overruns and unexpected hurdles, the author gained valuable lessons about careful model selection and the importance of tracking AI usage costs and environmental impact.