reasoning_effort

Provider name for _ChatOpenAICodex models authenticated via ChatGPT OAuth.

Distinct from "openai" (which uses an OPENAI_API_KEY) because the auth source, model class, and request endpoint all differ. See deepagents_code.integrations.openai_codex for the OAuth flow.

Return reasoning efforts supported by model_spec.

Returns plain str labels rather than EffortLabel: this is the public boundary where the label vocabulary is intentionally dropped, since the values flow straight to the UI.

Return the documented default reasoning effort when known.

Returns a plain str rather than EffortLabel: like supported_efforts_for_model, this is the public boundary where the label vocabulary is intentionally dropped, since every caller treats the value as display text.

Translate an effort label into provider-specific model params.

Read the configured effort from model params when present.

Merge effort params into existing per-session model params.

Remove known effort params while preserving unrelated model params.

A model specification in provider:model format.

Provider-specific reasoning effort behavior.

Closed vocabulary of effort labels across all supported providers.

Typing the per-provider tuples with this alias catches typos in the vocabulary at check time. It does not express the deeper invariant that a label must be supported by a specific model — that is enforced at runtime by supported_efforts_for_model.

This vocabulary is also hand-duplicated as display text in the /effort argument_hint (command_registry.py) and in COMMANDS.md; those are not type-checked against this alias, so update them in lockstep when it changes.

Provider identifiers that support model-specific reasoning effort controls.

Values must stay byte-identical to the provider strings from ModelSpec.parse used throughout model_config.py (e.g. CODEX_PROVIDER).