Frontier intelligence.
In your jurisdiction.

Build at the frontier with GLM 5.2

Frontier open-source models, hosted in Europe, never gated.
Point Claude Code at our base URL and cut your bill 80%.

Run your first model Calculate your savings

Built by experts from

Calculate your savings

See what your current Claude Code spend looks like running on Lyceum's near-frontier open models.

What you spend on Claude €5.000/mo

€5k€50k

Lyceum price €667/mo

Provider	Monthly cost	You save
GLM 5.2	€1.111/mo	78%
Kimi K2.6	€867/mo	83%
DeepSeek V4 Pro	€1.167/mo	77%
Qwen3.5 397B	€667/mo	87%

Every frontier open model, one endpoint.

Point your client at our base URL and call GLM, DeepSeek, Qwen, Llama, Kimi and more — all hosted in Europe, never gated.

GLM-5.2 Qwen3-235B-A22B Qwen3-32B Qwen3-30B-A3B Cosmos3-Super-Reasoner Nemotron-3-Nano-Omni Nemotron-Ultra-253B Nemotron-3-Nano-30B Llama-3.3-70B Kimi-K2.6 DeepSeek-V4-Pro Qwen3.5-397B-A17B MiniMax-M2.5 Nemotron-3-Ultra-550b Nemotron-3-Super-120b-a12b

Nemotron-3-Super-120b-a12b Nemotron-3-Ultra-550b MiniMax-M2.5 Qwen3.5-397B-A17B DeepSeek-V4-Pro Kimi-K2.6 Llama-3.3-70B Nemotron-3-Nano-30B Nemotron-Ultra-253B Nemotron-3-Nano-Omni Cosmos3-Super-Reasoner Qwen3-30B-A3B Qwen3-32B Qwen3-235B-A22B GLM-5.2

Your data never leaves Europe.

EU-hosted, zero-retention, and certified. Backed by real engineers.

Hosted in Europe

Every token runs on EU soil.

Zero data retention

Never stored. Never trained on.

ISO 27001

24/7 Support

Real engineers in Slack.

Everything's built in

Prompt caching, logs, traces, MCP, CLI.

The bottom line

lower bills on the same workloads

Everything you need to run AI workloads.

Inference APIs, serverless execution, GPU VMs, and managed clusters - designed to work individually or together.

Serverless Inference

Pay-per-token API access to open-source models

inference.py

from lyceum import Client

client = Client("your-api-key")

response = client.chat.completions.create(
    model="meta-llama/Llama-3-70B",
    messages=[{"role": "user", "content": "Hello!"}],
    stream=True
)

for chunk in response:
    print(chunk.choices[0].delta.content)

# Streaming response...

Hello! How can I assist you today?|

Dedicated Endpoints

Reserved GPU capacity for production models

Serverless Training

Run code on GPUs without managing infrastructure

Terminal

$ lyceum python run train.py -m gpu.h100

Provisioning H100... done

Uploading files... done

Epoch 1/10 | loss: 0.423 | acc: 0.891

Epoch 2/10 | loss: 0.312 | acc: 0.924

Epoch 3/10 | loss: 0.198 | acc: 0.957|

GPU Virtual Machines

Full root access, ready in seconds

Large-Scale Clusters

8 to 8,000 GPUs with InfiniBand

Active Queued Available 128 × H100 · 94% util

Built for teams that ship.

We handle the infrastructure complexity so you can focus on building. No capacity planning, no vendor lock-in, no surprises on your bill.

Per-second billing

Pay only for compute you actually use. Jobs that finish early don't cost you for time you didn't need.

Docker-native

Any Docker container runs on Lyceum with no modifications. No proprietary SDKs, no vendor lock-in.

EU data centres

GPUs hosted in European data centres. Full GDPR compliance, data residency in the EU.

Instant availability

GPUs provision in seconds, not hours. No capacity planning, no procurement queues.

38%

Shorter queue times

76%

Average GPU utilisation

<1m

Time to first job

99.9%

Platform uptime

GDPR Compliant · EU Data Residency

Transparent pricing.

Pay for what you use. No hidden fees.

GPU	VRAM	On-demand	12-mo commit	Save
NVIDIA B300	288 GB	$7.99/hr	$7.35/hr	−8%
NVIDIA B200	192 GB	$6.49/hr	$5.97/hr	−8%
NVIDIA H200	141 GB	$4.29/hr	$3.94/hr	−8%
NVIDIA H100	80 GB	$2.79/hr	$2.56/hr	−8%
NVIDIA A100	80 GB	$1.59/hr	$1.48/hr	−7%
NVIDIA L40S	48 GB	$1.19/hr	$1.10/hr	−8%

Per-second billing · cancel anytime on-demand

Frontier intelligence, hosted in Europe, never gated, at a fraction of the price.

Frontier intelligence. In your jurisdiction.