Run open models in production.
One OpenAI-compatible API. Per token, no base fee.
Run AI on European infrastructure.
Open models through one European API.
Dedicated GPUs when you need more control.
Teams building on Lyceum
References published as customers approve them.
Two product paths: inference and compute
Train and scale on European GPUs.
Virtual machines, training and clusters. Per second, no subscription.
Explore compute 03 / Our models One API.
All models and prices One API.
The right model.
- Language
GLM-5.3
1M context, tool use
$1.75 in · $4.50 out per 1M tokens
- Language
Kimi K3
Long context, agentic tool use
$3.00 in · $15.00 out per 1M tokens
- Language
GLM-5.3 Flash
Fast, low-cost, 1M context
$0.20 in · $0.50 out per 1M tokens
- Code
DeepSeek V4 Pro
Reasoning and coding
$1.75 in · $3.50 out per 1M tokens
- Code
Kimi K2.7 Code
Coding-focused agentic model
$1.25 in · $4.50 out per 1M tokens
- Multimodal
Kimi K2.6
Native multimodal, text and images
$1.00 in · $4.00 out per 1M tokens
- Language
Qwen3.8 2.4T A95B
Flagship MoE, function calling
$2.50 in · $6.00 out per 1M tokens
- Language
Qwen3.8 Flash Next
Preview of the Qwen 4 architecture
$0.20 in · $0.50 out per 1M tokens
- Language
MiniMax M3
1M context for large documents
$0.40 in · $2.00 out per 1M tokens
- Language
GLM-5.2
Bilingual reasoning, tool use
$1.50 in · $4.50 out per 1M tokens
- Language
gpt-oss 120B
Open-weight 120B
$0.15 in · $0.60 out per 1M tokens
- Language
Llama 3.3 70B
Instruction-following 70B
$0.13 in · $0.40 out per 1M tokens
- Embedding
Qwen3 Embedding 8B
Multilingual retrieval
$0.01 per 1M tokens
- Language
Qwen3.8 27B
Dense 27B, fine-tuning ready
$0.40 in · $2.40 out per 1M tokens
- Language
Qwen3.5 9B
Compact, fast, low cost
$0.15 in · $0.20 out per 1M tokens
- Language
Qwen3 235B A22B
Large MoE, instruction following
$0.20 in · $0.60 out per 1M tokens
- Language
Qwen3 30B A3B
Efficient MoE
$0.10 in · $0.30 out per 1M tokens
- Language
Qwen3 Next 80B Thinking
Reasoning variant
$0.15 in · $1.20 out per 1M tokens
- Multimodal
Gemma 3 27B
Instruction-tuned 27B
$0.10 in · $0.30 out per 1M tokens
- Language
Hermes 4 405B
Strong reasoning, long context
$1.00 in · $3.00 out per 1M tokens
- Language
Nemotron Ultra 253B
Ultra-large, Llama 3.1 based
$0.60 in · $1.80 out per 1M tokens
- Language
Nemotron 3 Nano 30B
Compact MoE, efficient
$0.06 in · $0.24 out per 1M tokens
- Multimodal
Nemotron 3 Nano Omni
Omni-modal reasoning for agents
$0.06 in · $0.24 out per 1M tokens
- Language
Cosmos 3 Super Reasoner
Multi-step reasoning
$0.10 in · $0.30 out per 1M tokens
- Multimodal OpenBMB
MiniCPM-V 4.5
Efficient vision-language
$0.66 in · $1.11 out per 1M tokens
04 / European sovereignty
Your AI stays close.
Only European data centres
Spain, Paris and the Nordics.
Zero data retention
Prompts and outputs are never kept or trained on.
One endpoint change
OpenAI-compatible API and plain Docker containers.
Hyper-available support
The people who run the platform, on the line.
Where we run and where we are
Data centre region Office
05 / GPU availability
Full pricing Choose the compute.
GPUOn-demandAvailabilityCommitment
-
B300
288 GB
On-demand$7.99 per GPU hour
Capacity Live in the dashboard → -
B200
192 GB
On-demand$6.49 per GPU hour
Capacity Live in the dashboard → -
H200
141 GB
On-demand$4.29 per GPU hour
Capacity Live in the dashboard → -
H100
80 GB
On-demand$2.79 per GPU hour
Capacity Live in the dashboard →
Reserved and dedicated capacity
From one server for one month to a dedicated cluster. Around 200 GPUs is the largest single-customer deployment running today. A 1,000-GPU deployment is in build.