Models

Listing models

GET https://api.artfical.com/v1/models
Authorization: Bearer tai-sk-...

Returns every model your key can use, in the same list shape OpenAI's own /v1/models endpoint uses:

{
  "object": "list",
  "data": [
    {
      "id": "tai-4.1",
      "object": "model",
      "created": 1749564000,
      "owned_by": "artfical",
      "name": "tAI 4.1",
      "description": "Our most capable model - top-tier reasoning and creativity",
      "context_length": 1000000
    }
    // ...
  ]
}

The models

Model	Positioning	Context
tai-4.0.6P	Fast, lightweight, for everyday use	128,000 tokens
tai-4.0.6B	Balanced: the middle ground between speed and quality	1,000,000 tokens
tai-4.0.7	Powerful, for complex tasks and deeper analysis	1,000,000 tokens
tai-4.1	Top-tier reasoning and creativity	1,000,000 tokens
tai-4.2	Latest generation, stronger reasoning and faster responses	200,000 tokens

Context windows differ by tier: tai-4.0.6P accepts up to 128,000 tokens, tai-4.2 up to 200,000, and tai-4.0.6B, tai-4.0.7 and tai-4.1 up to 1,000,000. The same values are returned as context_length by GET /v1/models. The window covers the prompt and the reply together. Every tier takes text input through the API; image input is not available yet, see Images. Per-model pricing is on the Pricing page.

Choosing a model

There's no single right answer here, but a reasonable default: start with tai-4.0.6P for anything latency-sensitive or high-volume, where "good enough, fast, and cheap" beats "as capable as possible." Move up to tai-4.0.7 or tai-4.1 when a task genuinely needs deeper reasoning, longer multi-step work, or very long inputs, and the extra cost and latency are worth it for that specific request. tai-4.2 is the newest tier and generally the best overall balance of capability and speed; reach for tai-4.1 specifically when a task benefits most from maximum reasoning depth over raw speed.

Nothing stops you from picking the model per-request rather than committing to one for an entire application. A common pattern is a fast model for routine turns and a more capable one invoked only when a request is flagged (by length, by keyword, by an earlier model's own assessment) as needing more depth.

An unrecognized model

Sending a model value that isn't one of the five above is rejected before any generation starts, with a 400:

{"error": {"message": "Unknown model: 'tai-4.9'. See https://docs.artfical.com/tai/en for the current list.", "type": "invalid_request_error"}}

This includes a typo'd model ID, an old ID that's been retired, or a model name from a different provider entirely (a request accidentally still pointed at a model string like "gpt-4" from a previous integration, for instance, since the base URL is what routes the request to tAI, but the model ID itself still has to be one of ours). See Errors for the full error shape.