Models

Model ids#

Ids are namespaced by vendor, anthropic/claude-sonnet-5, openai/gpt-5.5, zhipu/glm-5.3-flash-derisked, and are used exactly as listed, including case. A request naming an id that is not in the catalog, or not in your key’s allowlist, is refused before anything is billed.

Ids are stable. When a vendor renames a model upstream, the catalog id you integrated against keeps working. On the rare occasion we rename a catalog id ourselves, the old one becomes an alias and keeps resolving to the same model, so nothing you have deployed has to change. Each entry lists its retired ids in aliases, and a request that uses one is answered normally, with X-Ahura-Model and your usage record carrying the current id. zhipu/glm-5.3-flash-uncensored is an alias of zhipu/glm-5.3-flash-derisked for this reason.

List models#

Returns every active model your key can call: the public catalog plus any models private to your organization, such as your own fine-tuned adapters. The shape is OpenAI’s list shape with extra fields per entry.

curl https://api.ahurasense.com/v1/models \
  -H "Authorization: Bearer $AHURA_API_KEY"

{
  "object": "list",
  "data": [
    {
      "id": "anthropic/claude-haiku-4.5",
      "object": "model",
      "created": 0,
      "owned_by": "ahura",
      "display_name": "Claude Haiku 4.5",
      "description": "Fast tier. 200K context. High-volume light workloads.",
      "modality": "chat",
      "capabilities": {
        "tools": true,
        "vision": false,
        "json_mode": true,
        "streaming": true,
        "max_output": 64000,
        "context_window": 200000
      },
      "pricing": {
        "input_cents_per_mtok": 100,
        "cached_cents_per_mtok": 10,
        "output_cents_per_mtok": 500
      },
      "prices": {
        "currency": "USD",
        "unit": "per_million_tokens",
        "input": 1,
        "cached_input": 0.1,
        "output": 5,
        "discount": null,
        "effective_input": 1,
        "effective_cached_input": 0.1,
        "effective_output": 5
      },
      "off_peak": null,
      "featured": true,
      "aliases": []
    }
  ]
}

Prices in the response#

pricing is the raw catalog block, in cents per million tokens. prices is the same money in the unit people quote, dollars per million tokens, with the discounted rates worked out for you:

A rate the catalog does not publish is null, never 0. Prices are what you pay; what a serving backend charges us is not in this response and is not published anywhere.

Hosted models#

Most catalog models are served by partner backends. Some run on AhuraSense GPU infrastructure; the table marks them hosted, and responses from them carry X-Ahura-Routing: managed. Today these are the uncensored builds: the same open weights, served without the vendor’s refusal tuning.

Things that differ for hosted models:

• Reasoning is under your control: reasoning_effort, an exact thinking budget, or none at all. See Reasoning effort.
• The gateway’s response cache does not apply to them; every request reaches the model.
• If every replica is starting up, the gateway returns 503 instance_warming_up with Retry-After: 10. Retry, and the request goes through.
• They run on AhuraSense infrastructure end to end; no partner receives the prompt.

Catalog#

Prices are per million tokens. Context and max output are in tokens. Updated from the live catalog every few minutes.