Models

Start with auto unless Croatian-focused generation is the main requirement for your application.

At a glance

Model	Best for	Choose it when
`auto`	General chat, Q&A, coding, prototyping, and multilingual tasks	You want the recommended default
`hrllm`	Croatian-language writing and conversations	Croatian output quality is the main requirement

Model availability and routing can change. For the models available to your account at runtime, call GET /v1/models.

Use `auto` by default

auto lets the LLM.kiwi gateway choose an available model without requiring you to change your integration.

{
  "model": "auto",
  "messages": [
    { "role": "user", "content": "Explain APIs in simple language." }
  ]
}

Use it for:

General-purpose assistants
Questions and summarization
Coding and technical tasks
Multilingual applications
Prototypes and first integrations

Because routing may evolve, test important application behavior rather than depending on the identity of an underlying model.

Use `hrllm` for Croatian

hrllm is intended for Croatian-language generation.

{
  "model": "hrllm",
  "messages": [
    { "role": "user", "content": "Objasni što je API jednostavnim riječima." }
  ]
}

Good candidates include:

Croatian customer-support assistants
Croatian article drafts and rewriting
Croatian-language internal tools
Translation involving Croatian

For English-only or general tasks, begin with auto and compare outputs before switching.

List models programmatically

curl "https://api.llm.kiwi/v1/models" \
  -H "Authorization: Bearer sk_kiwi_..."

A typical OpenAI-compatible response contains a data array:

{
  "object": "list",
  "data": [
    { "id": "auto", "object": "model" },
    { "id": "hrllm", "object": "model" }
  ]
}

Use the model id exactly as returned by the API.

Make the choice

I am completely new

Use auto. It is the simplest starting point and is used throughout the quickstart.

My application must answer in Croatian

Test hrllm and auto with real examples from your application, then keep the option that produces more reliable results.

I am building a production application

Avoid assuming that auto always maps to the same underlying model. Validate structured outputs, set timeouts, and handle temporary errors.

I want to compare them fairly

Send the same system instruction and user message to both choices. Compare correctness, language quality, response time, and consistency across several realistic inputs.

Working examples

Copy examples for Python, JavaScript, cURL, and streaming.

Limits and pricing

Understand free usage and rate limits.

Get started

Build with LLM.kiwi

API reference

Help and resources

At a glance

Use `auto` by default

Use `hrllm` for Croatian

List models programmatically

Make the choice

Working examples

Limits and pricing

​At a glance

​Use auto by default

​Use hrllm for Croatian

​List models programmatically

​Make the choice

Working examples

Limits and pricing

At a glance

Use `auto` by default

Use `hrllm` for Croatian

List models programmatically

Make the choice