Choosing a Model
Available models and prices are always as shown live in the model plaza — the docs do not maintain a list. This page teaches you how to choose: first how to read the plaza, then which class of model fits which task.
How to read the model plaza
Each group card carries the information that matters when creating a key:
Each model card shows the model ID and context-tier prices (base tier / extended tier); "View detailed pricing" opens the full table including Fast Mode prices.
Think in tiers
Model families typically come in three tiers, and a card's description and pricing make its position obvious:
Rules of thumb:
- Get it working first, then tune the tier: start balanced, move up when quality falls short, move down when volume grows;
- Flagship and light tiers can differ by one to two orders of magnitude in unit price — bulk jobs on the wrong tier are the most common waste;
- Reasoning strength on thinking models is tunable via
reasoning_effort, see Advanced API Usage.
By task
Anthropic-protocol clients (such as Claude Code) should connect through an Anthropic-format group — see which format to use; live listings per format follow the model plaza.
Three things to check before switching models
- Context needs: long sessions crossing into the extended tier raise the unit price — not every task needs long context;
- Format match: models belong to groups and groups bind protocol formats; switching client protocol means switching groups (or use a composite key to hang several groups off one key);
- Verify before scaling: trial a new model on low traffic and check quality and cost in the usage logs before going all-in.
Related
- Billing — pricing formula, context tiers, charge order
- Core Concepts — groups / pools / multipliers
- Advanced API Usage — parameters and SDKs
- Fast Mode — the high-priority lane