TopolearnThe model is the biggest lever on quality, speed and price of every call you make.
Anthropic offers several models at once, from large and capable to small and fast. In October 2026 the docs list four current ones:
- Claude Opus (
claude-opus-5-5): the recommended default for most work, strong at long agent tasks and coding. - Claude Fable (
claude-fable-5-1): the most capable, for the hardest reasoning when Opus falls short. - Claude Sonnet (
claude-sonnet-5-5): a balance of speed and intelligence. - Claude Haiku (
claude-haiku-5-5): fastest and cheapest, for high volume jobs such as classifying, extracting fields or routing requests.
These names will change; new versions arrive every few months and old ones are retired on a published schedule. Read the models overview and the deprecations page before you pick, and keep the id in configuration so switching is a one-line change.
A practical way to choose: start with the default, build a small set of test questions, then try a cheaper model on the same set. If the cheaper one is good enough, use it. Larger models cost more per token and are slower, so the "best" model is often overkill for simple tasks.
You can also ask the API which models your key can use. client.models.list() returns each model's id, display name and, on current API versions, its limits and capabilities. Code that checks this list fails clearly instead of sending requests to a model that no longer exists.
Example
import anthropic
client = anthropic.Anthropic()
for m in client.models.list(limit=20):
print(m.id, "-", m.display_name)
# claude-opus-5-5 - Claude Opus 5.5
# claude-sonnet-5-5 - Claude Sonnet 5.5
# ... (your list depends on the date and your account)Remember
- Bigger models: more capable, slower, pricier; smaller: fast and cheap
- Start with the recommended default, then test cheaper models
- Model ids change; keep them in config and check the models list
Watch out for
- Hard-coding a model id across a codebase and discovering it was retired.