Available Models

maKI provides access to locally hosted models through an OpenAI-compatible API. All models run on dedicated GPU infrastructure at Uni Mannheim — your data never leaves the university network.

Loading model status …

This table is generated from the live status service. Model names and availability come from /docs/api/status.json.

The model list above is loaded live from the maKI status service. This keeps model names and availability in sync with the API without manually updating this documentation page.

Text-to-speech

Two TTS models are available via the OpenAI-compatible /v1/audio/speech endpoint. Important: each model only accepts its own voices — voice names from other APIs (e.g. alloy or cherry) are rejected.

Model	German voices	Other voices	Notes
voxtral-tts	de_male, de_female	neutral_male/female, casual_male/female, cheerful_female, fr_*, es_*, it_*, nl_*, pt_*, hi_*, ar_male	best German quality
qwen3-tts	— (voices are multilingual)	aiden, dylan, eric, ono_anna, ryan, serena, sohee, uncle_fu, vivian	pass "language": "German"; control style via "instructions"

curl https://maki.uni-mannheim.de/v1/audio/speech \
  -H "Authorization: Bearer ${MAKI_API_KEY}" \
  -H "Content-Type: application/json" \
  -d '{"model": "voxtral-tts", "input": "Hello from Mannheim.", "voice": "de_female"}' \
  --output hello.wav

Listing models

You can check which models are available for your API key at any time:

MAKI_API_KEY="..."
curl https://maki.uni-mannheim.de/v1/models \
  -H "Authorization: Bearer ${MAKI_API_KEY}"

Or with the OpenAI SDK:

from openai import OpenAI

client = OpenAI(
    base_url="https://maki.uni-mannheim.de/v1",
    api_key="your-api-key",
)

for model in client.models.list():
    print(model.id)