Dansk LLM

Models

Qwen3.8 27B

Alibaba's open model. Strong at code and tool calls, and it can think before it answers.

This model isn't running right now. Requests to it get 404 with model_not_found. See Models for the ones you can use.

Model id
qwen3.8-27b
Context window32,768 tokens (prompt and answer together)
InputText
OutputText
StreamingYes
Prompt cachingYes, automatic, in whole blocks of 1,568 tokens. Shorter prompts are not cached.
Tool calling (tools)Yes
Thinking (reasoning_effort)Yes, off unless you ask for it
PriceUnder Oversigt (Overview) on the platform
LicenseApache 2.0

Use the model

curl https://api.danskllm.dk/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DANSKLLM_API_KEY" \
  -d '{
    "model": "qwen3.8-27b",
    "messages": [
      {"role": "system", "content": "Du er en hjælpsom assistent."},
      {"role": "user", "content": "Forklar kort, hvad en API-nøgle er."}
    ]
  }'

Thinking

The model can think before it answers. That gives better answers on hard tasks, such as code, maths and planning, but the answer takes longer and costs more. You choose with reasoning_effort:

reasoning_effortMeans
Not set, or "none"No thinking. Fastest and cheapest.
"low"Brief thinking.
"medium"Ordinary thinking.
"high"Thorough thinking: the model checks its assumptions and considers other solutions. Slowest and most expensive.

Other values get status 400 with param: "reasoning_effort".

curl https://api.danskllm.dk/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $DANSKLLM_API_KEY" \
  -d '{
    "model": "qwen3.8-27b",
    "reasoning_effort": "low",
    "messages": [
      {"role": "user", "content": "Write a Python function that finds duplicates in a list."}
    ]
  }'

What to know about thinking:

  • It comes separately. The answer is in content as always. The thinking is in the reasoning field next to it, and in a stream in delta.reasoning. The field is not part of OpenAI's API, so the SDKs don't have it in their types, but it is in the response.
  • It is paid as output. Thinking counts towards completion_tokens.
  • It counts towards max_tokens. If you set max_tokens low, thinking can use all of it, so content is empty and finish_reason is "length". Set max_tokens high, or leave it unset, when you use thinking.
  • It takes time. With "high", thinking can run to several thousand tokens, so it can take a while before the answer starts. Use a stream if a user is waiting for the answer.
  • Send back only content. When the conversation continues, earlier answers need only content. If you send reasoning back, it becomes part of the prompt and costs input.

Coding agents and other tools

Tools that can use an OpenAI-compatible provider, such as Cline, Continue and opencode, are set up like this:

SettingValue
Base URLhttps://api.danskllm.dk/v1
API keyYour key from the platform, dkllm-...
Modelqwen3.8-27b
Context window32768
Reasoning effortnone or low for quick steps, high for hard tasks

Set the context window in the tool to 32,768. Then the tool shortens the conversation before it gets too long, instead of getting 400 with context_length_exceeded.

Tool calls work as with OpenAI: send tools, and the model answers with tool_calls. Send the result back as a message with role: "tool".

By default an account can have 4 requests running at once. See Limits and fair use if your tool sends more at the same time.

More about parameters and streaming.

On this page