Guides

Models

Find a model id, see what it costs, check where it’s hosted and what happens to your data, and how abliterated models work.

On this page

Find a model

The Models page lists every model with its prices, context window, maximum output, suggested tasks and data labels, and a button to copy its id. The same catalog is in your dashboard under Models. In code, GET /v1/models lists the ids your key can use.

Use an id exactly as it’s shown, such as deepseek-v4.1-flash-1017-raw. A model that isn’t in the catalog, or can’t be called right now, gets a 404 with the code model_not_found.

Prices

Each model has four prices, in US dollars per million tokens:

PriceCharged for
InputThe tokens you send: messages, instructions and tool definitions.
OutputThe tokens the model writes, including any reasoning.
Cache readInput the model reads from its prompt cache instead of processing again.
Cache writeInput written to the prompt cache, on models that charge for it.

Prices can change. A request is always charged at the prices in force when it ran, and the Usage page shows the prices each request used. See How a request is priced.

Reply length

  • Set the longest reply you want with max_tokens (max_output_tokens on Responses). If you leave it out on chat completions or Responses, the limit is 16,000 tokens. Messages always needs it.
  • Asking for more than a model can write gives you that model’s maximum, without an error. The Models page shows each model’s maximum, such as “16K max output”. The Anthropic-format model list gives the same number as max_tokens: see List models.
  • A reply that reaches the limit stops with finish_reason: "length", or stop_reason: "max_tokens" on Messages.
  • The context window on the Models page, such as “128K context”, is how much the model can read and write in one request. A request bigger than that can’t be answered and gets an error.

Before a reply starts, enough credit is set aside to cover the longest reply you allowed. A large max_tokens needs a larger balance, even if the reply turns out short. See Credit set aside before a reply.

Hosting and data labels

Under each model, the Models page shows what’s known about how your requests are handled:

LabelMeans
Hosted in: UK, EU, US or elsewhereWhere the model runs, and so where your request is processed.
No data kept by the model hostThe host doesn’t keep your prompts or replies after answering.
Model host keeps data up to N daysThe longest the host may keep them.
Not used for trainingThe host doesn’t train models on your requests.
Model host may train on dataThe host’s terms allow it to.

Each label shows the least favourable case for that model. Where something isn’t known, no label is shown rather than a guess. Your prompts pass through Kalsenti and the model host to produce a reply: see Data and privacy.

Models that need ID verification

Some models are marked ID verification required on the Models page. They only run for accounts that have passed an ID check. Until yours has, a request to one gets a 403 and is never answered by another model instead:

Response
{
  "error": {
    "message": "This model needs ID verification. Verify your identity in your Kalsenti dashboard, under Settings, then try again.",
    "type": "invalid_request_error",
    "param": null,
    "code": "identity_verification_required"
  }
}

To verify, go to Settings → Identity in your dashboard. The check takes a photo of your ID and a selfie, and needs a confirmed email address and at least one top-up first. If you can’t do a selfie check, email support@kalsenti.ai to ask for a manual review.