Models

The Arbitex Gateway exposes two OpenAI-compatible model listing endpoints. Both require a valid API key and return only the models that are active and that your organization is permitted to use.

Base URL: https://api.arbitex.ai/v1

Model identifier format

Every model is addressed using a provider/model-id compound string:

anthropic/claude-sonnet-4-20250514
openai/gpt-4o
google/gemini-2.0-flash

Pass this string in the model field of a chat completions request. The provider prefix routes the request to the correct upstream; the model-id is forwarded verbatim.

If you omit the provider prefix and send only a bare model-id (e.g. gpt-4o), the gateway attempts resolution against your configured provider list. This works for compatibility with existing integrations, but the provider/model-id format is preferred — it is unambiguous and avoids accidental routing mismatches when the same model-id exists on multiple providers.

A 404 model_not_found is returned when the resolved model is not recognized or is not available for your organization. See error codes for the full error body format.

ModelInfo object

Both listing endpoints return arrays of ModelInfo objects with the following fields:

Field	Type	Description
`model_id`	string	Provider-assigned model identifier (e.g. `gpt-4o`, `claude-sonnet-4-20250514`)
`display_name`	string	Human-readable model name
`provider`	string	Provider name (e.g. `openai`, `anthropic`, `google`)
`max_tokens`	integer	Maximum output tokens supported by this model
`supports_streaming`	boolean	Whether the model supports server-sent event streaming
`context_window`	integer \| null	Provider-reported maximum context window in tokens; `null` when not exposed by the provider’s catalog
`capabilities`	array of string	Capability tags — values include `chat`, `vision`, `tool_use`, `streaming`

The grouped endpoint adds three computed fields per model entry (see List models grouped):

Field	Type	Description
`tier`	string	Model tier inferred from the model-id naming convention (e.g. `opus`, `sonnet`, `pro`)
`version_score`	array of integer	Extracted version digits used for ordering (e.g. `[4, 6]` for `claude-sonnet-4-6`)
`is_latest_in_tier`	boolean	Whether this is the highest-version model in its `(provider, tier)` bucket

List models

GET /v1/models

Returns a flat array of all active models available to the calling user. Provider model listings are cached for up to one hour; admin-managed availability flags are checked fresh on every request. Only text-chat-capable models are included — image, audio, TTS, and embedding models are excluded.

Access control is applied: if your organization has model access rules configured (allowlists or denylists by provider or model pattern), only models the calling user is permitted to use are returned. See model access rules for configuration details.

Request

curl https://api.arbitex.ai/v1/models \
  -H "Authorization: Bearer $ARBITEX_API_KEY"

Response 200 OK

Cache-Control: public, max-age=3600

[
  {
    "model_id": "claude-sonnet-4-20250514",
    "display_name": "Claude Sonnet 4",
    "provider": "anthropic",
    "max_tokens": 8192,
    "supports_streaming": true,
    "context_window": 200000,
    "capabilities": ["chat", "vision", "tool_use", "streaming"]
  },
  {
    "model_id": "gpt-4o",
    "display_name": "GPT-4o",
    "provider": "openai",
    "max_tokens": 16384,
    "supports_streaming": true,
    "context_window": 128000,
    "capabilities": ["chat", "vision", "tool_use", "streaming"]
  },
  {
    "model_id": "gemini-2.0-flash",
    "display_name": "Gemini 2.0 Flash",
    "provider": "google",
    "max_tokens": 8192,
    "supports_streaming": true,
    "context_window": 1048576,
    "capabilities": ["chat", "vision", "streaming"]
  }
]

List models grouped

GET /v1/models/grouped

Returns the same model set as GET /v1/models but reshaped for a model picker UI. The highest-version model per (provider, tier) is placed in current; older versions appear in legacy. Tier is inferred from the model-id naming convention — for example, claude-sonnet-4-20250514 is tier sonnet, gpt-4o is tier gpt-4o, gemini-2.5-pro is tier pro.

All the same access control filtering and activity-flag filtering as the flat endpoint applies.

Request

curl https://api.arbitex.ai/v1/models/grouped \
  -H "Authorization: Bearer $ARBITEX_API_KEY"

Response 200 OK

Cache-Control: private, max-age=60

{
  "current": [
    {
      "provider": "anthropic",
      "models": [
        {
          "model_id": "claude-opus-4-5",
          "display_name": "Claude Opus 4.5",
          "provider": "anthropic",
          "max_tokens": 8192,
          "supports_streaming": true,
          "context_window": 200000,
          "capabilities": ["chat", "vision", "tool_use", "streaming"],
          "tier": "opus",
          "version_score": [4, 5],
          "is_latest_in_tier": true
        },
        {
          "model_id": "claude-sonnet-4-20250514",
          "display_name": "Claude Sonnet 4",
          "provider": "anthropic",
          "max_tokens": 8192,
          "supports_streaming": true,
          "context_window": 200000,
          "capabilities": ["chat", "vision", "tool_use", "streaming"],
          "tier": "sonnet",
          "version_score": [4, 0],
          "is_latest_in_tier": true
        }
      ]
    },
    {
      "provider": "openai",
      "models": [
        {
          "model_id": "gpt-4o",
          "display_name": "GPT-4o",
          "provider": "openai",
          "max_tokens": 16384,
          "supports_streaming": true,
          "context_window": 128000,
          "capabilities": ["chat", "vision", "tool_use", "streaming"],
          "tier": "gpt-4o",
          "version_score": [4, 0],
          "is_latest_in_tier": true
        }
      ]
    }
  ],
  "legacy": [
    {
      "provider": "anthropic",
      "models": [
        {
          "model_id": "claude-sonnet-4-5",
          "display_name": "Claude Sonnet 4.5",
          "provider": "anthropic",
          "max_tokens": 8192,
          "supports_streaming": true,
          "context_window": 200000,
          "capabilities": ["chat", "vision", "tool_use", "streaming"],
          "tier": "sonnet",
          "version_score": [4, 5],
          "is_latest_in_tier": false
        }
      ]
    }
  ],
  "pinned_legacy": []
}

Response structure

Field	Type	Description
`current`	array	Provider groups containing the latest model per tier; sorted alphabetically by provider name
`current[].provider`	string	Provider name
`current[].models`	array	ModelInfo objects (with `tier`, `version_score`, `is_latest_in_tier`), sorted alphabetically by tier name
`legacy`	array	Provider groups for older model versions; models sorted version-score descending (newest first)
`pinned_legacy`	array	Flat list of tenant-pinned legacy models; these also appear in `current` with `is_latest_in_tier: true`

Provider routing

When the gateway receives a chat completions request it resolves the provider from the provider/model-id string, then forwards the request to the upstream API using the provider credentials configured for your organization.

If multiple provider credentials are configured for the same provider (for example, two Anthropic keys with different rate limits), the gateway selects among them according to your routing policy. If the selected upstream is unavailable, the gateway follows your configured failover chain. A 503 provider_unavailable is returned only when all providers in the chain are exhausted.

Error responses

Status	Code	Description
`401`	`invalid_api_key`	API key not found, revoked, or malformed
`404`	`model_not_found`	The requested `provider/model-id` is not recognized or not available for your organization