Running Claude Code subagents on Opus instead of Sonnet cost $1,600 in two months
Sankra · reddit · 2026-11-20
After hitting his limit once too often, the author audited two months of Claude Code logs. Subagents consumed 55% of all tokens, and 89% of that ran on Opus — setting model: sonnet in the subagent definition would have cut about $1,600 at API rates.
Other findings:
- Cache misses after breaks added up to $270
- One 55-hour session cost $93
- Two token-saving plugins combined saved only $158
He published ten token-saving changes with the numbers behind each on his blog.
More from coding & agent
- CapCut plugin connects ChatGPT to auto-cut gameplay highlights, then human picks the payoff — iamfakhrealam · 2026-11-21
- Dev wires a Runway-generated avatar into Nous Research's Hermes Agent, bashing boring agent UIs — alexcovo_eth · 2026-11-21
- Lean compiles, but the natural-language proof is still wrong — jasondeanlee · 2026-11-21
- DeepMind demos ski simulator built by Gemini 4 Neon from real Mont Blanc terrain — GoogleDeepMind · 2026-11-21
- pstack v0.15.46 trims skills, adds Gemini as panel reviewer; pinning question raised — petrusenko_max · 2026-11-21
- Hamel Husain on evals: a good process beats good tools — HamelHusain · 2026-11-21