Models
Open models, run by us or by European partners.
Open models of the current generation, including Gemma 4, Qwen 3.8 and GLM 5.3, mostly under Apache 2.0 or MIT licences. At our partners they answer without wait time, and you choose per chat. Here you can see which model suits which task and where it runs.
Fast
Gemma 4 12B with 262k tokens of context: several hundred pages in one question. For everyday work.
Balanced
Qwen 3.8 27B for reports, analysis and longer correspondence. Run by us, or without cold start in the partner data centre.
Strong
Qwen 3.8 Max and Kimi K3 in the partner data centre, GPT-OSS 120B at STACKIT in Germany. For long texts and difficult matters.
Reasoning
Shows the reasoning separately from the answer. You set how thoroughly the model thinks in up to three levels.
Vision
Gemma 4 understands images: sketches, floor plans, forms, photos.
Coding
GLM 5.3 without cold start and Qwen 3.6 35B-A3B run by us, for Cline, Continue, Zed and Aider.
Also in the catalogue
Embeddings
Make texts searchable and power the knowledge base. Multilingual (BGE-M3).
Re-ranker
Sorts search hits by relevance to the question (BGE reranker).
Read aloud
German and international voices in nine languages, generated in EU data centres. Samples in the voice catalogue.
Dictation
Speak into the microphone and the text appears in the input field (Voxtral, EU endpoint).
Decision Engine
Probabilities for fixed options, in milliseconds.
All models you can choose in the chat today
This list comes straight from the product's model catalogue — the same source that feeds the model picker and the API. It is therefore always up to date, not a maintained copy.
Run by us5 models
Runs on hardware we rent and configure ourselves — in data centres in Romania, Sweden, Czechia, the Netherlands and France, as well as in Iceland and Norway (EEA).
Gemma 4 12B
Standard
Google DeepMind · US · Apache 2.0
run by us in European data centres · Powered by Gemma
Qwen 3.8 27B
Allrounder
Alibaba · CN · Apache 2.0
run by us in European data centres
Tiefdenker · 9B
Reasoning
Jackrong (Community-Distill) · CN · Apache 2.0
run by us in European data centres
Gemma 4 26B-A4B (Vision)
Vision
Google DeepMind · US · Apache 2.0
run by us in European data centres · Powered by Gemma
Coder mini · 35B-A3B
Coder · schnell
Alibaba · CN · Apache 2.0
run by us in European data centres
Run by partners15 models
Runs at European partners with contractually guaranteed zero retention of inputs and outputs. No cold start, but third-party hardware.
Qwen 3.6 27B · STACKIT
Allrounder · Deutschland
Alibaba · CN · Apache 2.0
STACKIT (Schwarz Digital), Deutschland
GPT-OSS 120B · STACKIT
Gründlich · regelbar
OpenAI · US · Apache 2.0
STACKIT (Schwarz Digital), Deutschland
Gemma 4 31B · STACKIT
Vision · Deutschland
Google DeepMind · US · Apache 2.0
STACKIT (Schwarz Digital), Deutschland
Qwen 3.8 Max
Bestes Ergebnis
Alibaba · CN · Apache 2.0
europäisches Partner-Rechenzentrum (Irland/Finnland)
Qwen 3.8 27B · ohne Kaltstart
Allrounder · sofort
Alibaba · CN · Apache 2.0
europäisches Partner-Rechenzentrum (Irland/Finnland)
Kimi K3
Gründlich
Moonshot AI · CN · Modified MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
DeepSeek V4 Flash
Schnell & günstig
DeepSeek · CN · MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
DeepSeek V4 Pro
Stark & sofort
DeepSeek · CN · MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
Kimi K2.7 Code
Coder · sofort
Moonshot AI · CN · Modified MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
MiniMax M3
Gründlich · günstig
MiniMax · CN · Lizenz ungeprüft
europäisches Partner-Rechenzentrum (Irland/Finnland)
GLM 5.2
Vielseitig · sofort
Z.ai · CN · MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
GLM 5.3
Bester Code
Z.ai · CN · GLM-5.3-Lizenz (MIT-artig)
europäisches Partner-Rechenzentrum (Irland/Finnland)
Qwen 3.8 Flash
Alltag · sparsam
Alibaba · CN · Qwen Community 1.0
europäisches Partner-Rechenzentrum (Irland/Finnland)
DeepSeek V4.1 Flash
Stark & günstig
DeepSeek · CN · MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
GLM 5.3 Flash
Günstig · sofort
Z.ai · CN · MIT
europäisches Partner-Rechenzentrum (Irland/Finnland)
You can also connect your own models (your own endpoint or your own provider key). They then appear only in your organisation and are not part of this list.
Decision models
The Decision Engine is selected as a version, not as a base model. Which runtime does the computing is set by the platform — the region is part of every response.
SovrGPT Decision Engine v1Default
sovr-decision-v1
Operation: Germany
Images: yes
SovrGPT Decision Engine v2
sovr-decision-v2
Operation: EU (Ireland/Finland)
Images: yes