Models

Models are the LLM options available in Aegis. Each has a URL-safe slug (what you pass in model_slug on runs and evaluations), a display name, and supplier metadata. Thinking models may also accept a reasoning_level on evaluate and run creation — allowed values are per model (see tables below). When you omit model_slug on those endpoints (or leave the dashboard selector unchanged), the default is gpt-5.6-luna with reasoning level none. For LLM-backed metrics, that model — or whichever model you select — is used for every model call in the evaluation. Aegis does not substitute a different model for individual steps.

For Factfulness and Plan Quality, a higher-capability model than the default is recommended for optimal results.

There is no GET /models (or related) route on the API server—the catalog is documented here.

A slug that has been deactivated still appears in this catalog (historical runs keep the slug they were created with) but returns 400 on evaluate and run creation (The following models are not active: …).

OpenAI

slug	name	thinking	latest	reasoning_levels	default_reasoning_level
gpt-4o	GPT-4o	no	no	—	—
gpt-4.1	GPT-4.1	no	no	—	—
o3-mini	o3-mini	yes	no	—	—
o4-mini	o4-mini	yes	no	—	—
gpt-5	GPT-5	yes	no	minimal, low, medium, high	medium
gpt-5.1	GPT-5.1	yes	no	none, low, medium, high	none
gpt-5.2	GPT-5.2	yes	no	none, low, medium, high, xhigh	none
gpt-5.4	GPT-5.4	yes	no	none, low, medium, high, xhigh	none
gpt-5.5	GPT-5.5	yes	no	none, low, medium, high, xhigh	none
gpt-5.6-luna	GPT-5.6 Luna	yes	no	none, low, medium, high, xhigh, max	none
gpt-5.6-terra	GPT-5.6 Terra	yes	no	none, low, medium, high, xhigh, max	none
gpt-5.6-sol	GPT-5.6 Sol	yes	no	none, low, medium, high, xhigh, max	none
gpt-6-astra	GPT-6 Astra	yes	yes	none, low, medium, high, xhigh, max	none

Anthropic

slug	name	thinking	latest	reasoning_levels	default_reasoning_level
claude-haiku-4-5	Claude Haiku 4.5	no	no	—	—
claude-sonnet-4-6	Claude Sonnet 4.6	yes	no	low, medium, high, max	high
claude-sonnet-5	Claude Sonnet 5	yes	no	low, medium, high, max	high
claude-opus-4-5	Claude Opus 4.5	yes	no	—	—
claude-opus-4-6	Claude Opus 4.6	yes	no	low, medium, high, max	high
claude-opus-4-7	Claude Opus 4.7	yes	no	low, medium, high, xhigh, max	high
claude-opus-4-8	Claude Opus 4.8	yes	no	low, medium, high, xhigh, max	high
claude-opus-5	Claude Opus 5	yes	no	low, medium, high, max	high
claude-fable-5	Claude Fable 5	yes	no	low, medium, high, max	high
claude-fable-5-1	Claude Fable 5.1	yes	yes	none, low, medium, high, xhigh, max	high

• thinking — yes means the model is treated as a “thinking” / reasoning-style model in the product; no means it is not.
• latest — yes marks the current recommended variant within that supplier’s line where applicable; other rows show no. A supplier may have more than one latest model (for example a fast and a flagship option).
• reasoning_levels — allowed values for reasoning_level on evaluate and run endpoints. — means the model does not accept a reasoning level (omit the field; supplying one returns 400).
• default_reasoning_level — value used when reasoning_level is omitted on a model that supports it. — when not applicable.

In API payloads, always use the slug (for example gpt-5.6-luna, gpt-6-astra, or claude-fable-5-1).