evroc Think
evroc Think models provide sovereign access to best-in-class open models, operated entirely in Europe. Developers get model access on shared or dedicated GPUs, easy access to AI they can run, adapt, and scale fast.
Why evroc.
Get the highlights
Frontier open models
Chat, code, embedding, transcription, and vision models in one catalogue.
OpenAI-compatible endpoints
Use the OpenAI API, standard clients, and LangChain with existing code.
Dedicated GPU instances
Your own isolated model instance, billed by the hour.
Open models.
Best in class AI models
Gemma-4-26B
GoogleChatCodeVision
€0.125/0.5/1M tokens (in/out)
multilingual-e5-large-instruct
Hugging FaceEmbedding
€0.1/0.1/1M tokens (in/out)
kb-whisper-large
KBLabTranscription
€2/1K mins
pianissimo-sv
Klang AITranscription
€2/1K mins
Mistral-Medium-3.5-128B
Mistral AIChatCodeVision
€1.5/6/1M tokens (in/out)
Kimi-K2.6
Moonshot AIChatVision
€1.25/5/1M tokens (in/out)
Llama-3.3-70B-Instruct
NvidiaChat
€1/1/1M tokens (in/out)
gpt-oss-120b
OpenAIChat
€0.2/0.8/1M tokens (in/out)
whisper-large-v3
OpenAITranscription
€2/1K mins
whisper-large-v3-turbo
OpenAITranscription
€2/1K mins
Qwen3-Embedding-8B
QwenEmbedding
€0.1/0.1/1M tokens (in/out)
Qwen3-Reranker-4B
QwenEmbedding
€0.05/1M tokens (input only)
Qwen3.6-35B-A3B
QwenChatCodeVision
€0.3/1.2/1M tokens (in/out)
Qwen3.8-27B
QwenChatCode
€0.75/3/1M tokens (in/out)
GLM-5.3
Zhipu AIChatCode
€1.25/5/1M tokens (in/out)
Highest security and performance where it matters most.
More results.
Building with evroc
AI costs cut by 3x
Compared to GPT-5 while maintaining output quality.
30x faster LLM inference
NVIDIA B200 compared to the NVIDIA H100 chip.
Transcribing with speed
Significantly faster than major hyperscalers.
