Tag · 9 notes
LLM
Notes on language models in coding agents, decision models, routing a request and choosing a model and effort level for a task.
2026
Does the strongest model pay off?In 1,961 headless coding runs, the strongest Claude pair passed most, but it cost far more per pass than cheaper pairs that passed nearly as often.
write-upWhere should a decision model decide?A model choice can happen per request, turn, subagent, session, or after failures. Each timing trades cache reuse for precision.
write-upShould a decision model pick your model?Letting a decision model pick the model for each coding task saves a choice, but it has costs. The reasons for it, against it, and three alternatives.
write-upRead a coding agent's model list without logging inCodex and Claude Code both ship their model picker data inside the installed package. You can read it without an account.
tilIs the cheaper model enough?Why a model recommender must sometimes try the cheaper model on purpose, and why a random control group makes the comparison fair.
write-upBuilding a model recommenderA model recommender in four steps and a feedback loop, why it can only recommend, and why linking a result to its task is a guess.
write-upWhat is a decision model?A decision model answers a fixed question with one of your options and a probability. That makes it fast and easy to check, if the options are right.
write-upRoute one Claude Code request to another modelClaude Code hooks can watch a session, but a mod can route one model request elsewhere and change its effort.
tilWho picks the model?Coding agents let you choose a model and an effort level. Why both easy answers cost something, and the idea of a model that picks for you.
write-up