• Sources: primary, discussion
  • Summary: Of 2B tokens for the month, 1B went to models other than the GLM 5.3 Flash target, and total energy came to about 35 kWh against a 10 kWh goal. A vibe-coded Wagtail MCP prototype consumed 450M tokens, $150, and 5 kWh almost overnight on what the author calls the wrong model choice, against $68 for the whole GLM 5.3 Flash share. The switch away was forced by capacity degradation at the inference provider rather than model quality, moving work to DeepSeek V4.1 Flash and Qwen 3.8 Flash.
  • Why it matters: A core team publishing per-model token, dollar, and energy figures gives a cost reference point for agentic coding that capability benchmarks do not provide.