Opus: deep reasoning
The most capable family. Use it for hard problems: multi-step analysis, tricky code, strategy, long documents where nuance matters. It is slower and costs more per message.
Lesson 2
Three model families, three trade-offs between depth, speed and cost. Pick the model to fit the job.
The most capable family. Use it for hard problems: multi-step analysis, tricky code, strategy, long documents where nuance matters. It is slower and costs more per message.
The balanced default. Strong enough for most writing, coding, analysis and research, fast enough to feel interactive. If you are unsure, start here.
The quickest and cheapest. Ideal for classification, extraction, short rewrites, high-volume jobs and anything where speed beats depth.
Start with Sonnet. If the answer is shallow or misses steps, move up to Opus. If you are running the same simple task hundreds of times, move down to Haiku. Extended thinking (when available) gives any model more room to reason before answering.
Version numbers move forward regularly. The pattern of a deep, a balanced and a fast tier has stayed stable, so learn the roles rather than the version numbers.
Before answering, tell me whether this task needs deep reasoning or is a quick lookup, and why: [paste your task]