Models
A model id encodes three things: provider, model family, and reasoning effort. The
format is provider:model__effort, for example anthropic:opus__xhigh or
openai:gpt-5.6-sol__high. The effort is optional, so anthropic:sonnet is also valid.
| Provider | Models | Reasoning efforts | Context window |
|---|---|---|---|
anthropic | fable, opus | low, medium, high, xhigh, max, ultracode | 1,000,000 |
anthropic | sonnet | low, medium, high | 1,000,000 |
anthropic | haiku | none | 200,000 |
openai | gpt-5.6-sol, gpt-5.6-terra | low, medium, high, xhigh, ultra | 1,050,000 |
openai | gpt-5.6-luna | low, medium, high, xhigh | 1,050,000 |
openai | gpt-5.5, gpt-5.4-mini | low, medium, high, xhigh | 1,050,000 and 400,000 |
A model is chosen per session. Omit it and the session runs on the workspace default.
A workspace authenticates with a provider one of five ways. Anthropic: api_key,
aws_bedrock, or claude_code. OpenAI: api_key or codex. Credentials for several
methods can be stored at once, but exactly one is active per provider. The OAuth methods
are logins that expire and refresh on a schedule, not static keys, and they are set up
through a short-lived auth sandbox.