balanced tier in Anthropic's lineup. Here's when the price actually makes sense, and when you're better off downgrading.
Most developer workloads. Code generation, content drafting, agent reasoning. The default choice if you don't know.
Simple classification, intent detection, short summaries. Haiku is 4x cheaper and often sufficient.
Most teams running Claude Sonnet 24/7 are over-using the tier. Audit reveals which requests actually needed this model and which would have worked on the tier below. Run the free audit →
Audit my Claude usage →