Open-weight LLMs, hosted in Austria

Every model with its own page: what it can do, how fast it runs here, how well it reasons and what it costs.

Logic Index and speed are measured by us. Click a model to open its page.

Decision models

They do not generate text. They receive a situation and typed questions and return a probability for each answer option, in one pass and via their own endpoint. Typical use: deciding which model gets a request.

How to call them

Our own models

Trained by us, run on our own hardware, open on Hugging Face. Small and specialised: as good as large models at their task, but in milliseconds instead of seconds.

Model	Task	Use	Duration	Price	Status
ALLSmartIntent 0.3B · mmBERT	decision per message	API /v1/intentHugging Face ↗	≈ 15 msper message	freewith any API key	loaded
ALLSmartArt4 0.3B · mmBERT	names, companies, addresses (Art. 4 GDPR)	privacy layer, from level maskHugging Face ↗	≈ 41 msper request	€ 0.10per M input tokens	loaded
ALLSmartArt4i 0.3B · mmBERT	people identifiable without a name (Art. 4 GDPR, indirect)	privacy layer, level strictHugging Face ↗	≈ 15 msper text	€ 0.10per M input tokens	loaded
ALLSmartArt9-10 4.2B + 0.3B · Qwen3.5-4B (LoRA) + mmBERT	health, belief, offences (Art. 9 and 10 GDPR)	privacy layer, special-category levelsHugging Face ↗	≈ 90 msper section	€ 0.10per M input tokens	loaded

The privacy price applies once per request with a privacy level, for all privacy models together. For licensing reasons, the public versions of the three privacy models on Hugging Face are separate trainings, not the same weights as here; ALLSmartIntent is the same model there. Accuracy, comparisons with large models and limits are on the model pages.

curl https://allsmartllm.com/v1/chat/completions \
  -H "Authorization: Bearer $KEY" \
  -d '{"model": "top", "messages": [ … ]}'

# another model: only the name changes
  -d '{"model": "coder", "messages": [ … ]}'

Own models

Trained in-house, measured blind.

Our own models detect personal data and special categories and thereby power the privacy layer. With all measurements and comparisons.

See own models