Differentiate between Claude model types (Haiku, Sonnet, Opus)

Key points

1. 1

   Haiku is the fastest, lightest tier: suited to quick lookups, simple edits, classification and high-volume straightforward tasks where speed and efficiency matter more than reasoning depth.

2. 2

   Sonnet is the balanced everyday model: fast and highly capable, designed for the vast majority of writing, analysis, question-answering and coding-help tasks. It is the sensible default.

3. 3

   Opus is the most capable tier for genuinely complex, multi-step problems with interacting constraints, used when quality matters more than speed or volume.

4. 4

   Describe the tiers by trade-off (speed and efficiency versus reasoning depth), not by price; exam items avoid pricing specifics.

5. 5

   Switch models by clicking the model name next to the send button. Switching mid-conversation keeps the history; the new model sees everything the previous one produced, and the change applies from the next response.

6. 6

   The picker remembers your last selection by default, and organisations can set a default model; a default is a convenience, not a recommendation for a given task.

7. 7

   Effort and extended thinking are settings, not tiers: they make a model work harder on a response (more thorough, slower, more usage). They do not turn the lightest tier into the best reasoner.

8. 8

   No tier is infallible. A more capable model reduces error rates but every model can produce plausible wrong details, so verification is still required.

9. 9

   Haiku follows instructions and produces structured output (tables, JSON-like lists); its limitation is depth on hard problems, not formatting.

10. 10

    Model tier does not govern memory across chats; that comes from the memory feature, Projects and chat search.

11. 11

    Common distractors: reversing the tiers; "Sonnet is only for code"; "bigger model remembers more between chats"; one bad result on a hard task used to justify moving all work to the heaviest tier.

Read the source

• Change the model, effort, and thinking settings
• Set a default model for your organization
• Models overview (Anthropic docs)