AALLM Docs

Models & deployments

Separate what a lab built from the concrete provider surface your application can actually call.

Identity model

model family
A long-lived product line, such as anthropic/claude-haiku.
model release
A canonical release produced by a lab, such as anthropic/claude-haiku-4.5.
deployment
A provider-specific model ID, region, API surface, limits, capabilities, and rate card.
rate card
Observed prices for one deployment, including their source and observation time.

Use deployment IDs for decisions

A canonical model can differ by provider, region, API surface, capability support, and price. Quote and lock the deployment—not only the model name.

Deployment shape

JSON
{
  "id": "dep_openai_gpt_5_1_responses_global",
  "model_id": "openai/gpt-5.1",
  "provider_id": "openai",
  "provider_model_id": "gpt-5.1",
  "api_surface": "responses",
  "mode": "responses",
  "region": "global",
  "status": "available",
  "lifecycle": {
    "status": "available",
    "announced_at": null,
    "deprecation": {
      "state": "unknown",
      "date": null,
      "precision": "unknown",
      "date_type": "unknown",
      "evidence": []
    },
    "end_of_life": {
      "state": "unknown",
      "date": null,
      "precision": "unknown",
      "date_type": "unknown",
      "evidence": []
    },
    "replacement_ids": [],
    "coverage": {
      "status": "partial",
      "last_checked_at": null,
      "source_count": 1,
      "source_ids": ["20000000-0000-0000-0000-000000000001"]
    },
    "last_changed_at": null
  },
  "context_tokens": 400000,
  "max_output_tokens": 128000,
  "capabilities": {
    "tool_calling": {
      "status": "supported",
      "verification_status": "official_declared"
    },
    "structured_output": {
      "status": "supported",
      "verification_status": "official_declared"
    }
  },
  "rate_card": {
    "currency": "USD",
    "unit": "million_tokens",
    "input_micro_usd": 1250000,
    "cached_input_micro_usd": 125000,
    "output_micro_usd": 10000000,
    "source_url": "https://developers.openai.com/api/docs/models/gpt-5.1",
    "observed_at": "2026-07-14T00:00:00Z"
  },
  "evidence_urls": [
    "https://developers.openai.com/api/docs/models/gpt-5.1"
  ]
}

Select a deployment

  1. Filter by canonical model or search term.
  2. Require the capabilities your workflow cannot operate without.
  3. Constrain provider, region, and mode when compliance or integration requires it.
  4. Compare only compatible deployment IDs using your expected token usage.
  5. Persist the selected ID with the catalog release used for the decision.

Unknown values are not zero

A null limit, missing rate card, or unknown capability means ALLM cannot assert the value from current evidence. Do not treat it as unlimited, free, or supported.