Models
Language models and decision models.
Each model gets its own card with the real numbers: pricing, context, quantisation, where it runs, and performance we measured ourselves rather than quoted from a spec sheet. New models are announced in the changelog and appear in GET /v1/models the same day, decision models only for a key that has them enabled.
Qwen3.8-27B
nvidia/Qwen3.8-27B-NVFP4 · id in the API:
nvidia/Qwen3.8-27B-NVFP4context262,144 tokens
max output32,768 tokens
inputtext and images
quantisationNVFP4 on NVIDIA RTX PRO 6000 Blackwell
pricing$0.25 / $2.09 / $0.04 per M (input / output / cached input)
runs inItaly (EU), edge in Nuremberg, Germany
data handlingzero retention of request and response content
Full model card with measured performance and GDPR answers →
Decider decision models
Mapika/decider, quantised by LLM Tech · ids in the API:
decider-4b-nvfp4, decider-2b-fp8, decider-0.8b-fp8endpointPOST /v1/systemone
stateup to 32,768 tokens fits with any question of up to 4,096
outputanswers to your questions, no text generation, 0 output tokens
quantisationNVFP4 (4b), FP8 (2b, 0.8b)
pricing$0.04 / $0.03 / $0.02 per M input (4b / 2b / 0.8b), output free
runs inItaly (EU), edge in Nuremberg, Germany
data handlingzero retention of request and response content
Programmatic list
The authoritative list is the API itself: GET https://api.llmtech.eu/v1/models. Without a key it returns the chat models with their context length, features and per-token prices; decision models are listed only for a key that has them enabled. Catalogue integrations should read that endpoint rather than this page.