Lyceum

Featured

EU-hosted inference cloud for open-source models, OpenAI-compatible

Lyceum is an inference cloud for open-source models, built and hosted in the EU. It offers serverless inference with pay-per-token, OpenAI-compatible API access, and dedicated endpoints for reserved capacity.

For training and larger jobs it also provides on-demand GPU VMs (1-8 GPUs, ready in seconds), large-scale clusters (8-8000 GPUs with InfiniBand), and serverless Python and Docker execution.

Data stays in European data centres, so teams with EU data-residency or sovereignty needs can run inference without leaving EU jurisdiction. Lyceum is Berlin-based (Germany).

Pricing: Usage-based

Hosting Cloud

Pricing From $0.06/1M tokens

HQ 🇩🇪 Germany

Founded 2025

Compliance GDPR

Visit website →

Pricing

GitHub

3 developers want to try this

What is Lyceum?

Lyceum is an inference cloud for open-source models, built and hosted in Europe. It brings serverless inference, dedicated endpoints, on-demand GPU VMs, and large-scale training clusters into one platform, with GPUs running in European data centres for teams that need EU data residency.

What it offers

Serverless inference: pay-per-token, OpenAI-compatible API access to open-source models including gpt-oss 120B, Qwen3.5, Kimi K2.6, GLM-5.2, MiniMax M3, and DeepSeek V4. Point an OpenAI-style client at Lyceum and stream completions.
Dedicated endpoints: reserved GPU capacity for production models when you need consistent throughput.
On-demand GPU VMs: full root access, provisioned in seconds, with per-second billing and no minimum commitment. Hardware runs from L40S and A100 up to H100, H200, and the Blackwell B200 and B300.
Large-scale clusters: 8 to 8,000 GPUs connected over InfiniBand for distributed training.
Serverless execution: run any Docker container or Python job on GPUs without managing infrastructure.

EU hosting and sovereignty

GPUs are hosted in European data centres with GDPR compliance and EU data residency, so inference and training stay under EU jurisdiction. This makes Lyceum a fit for teams with sovereignty, residency, or compliance requirements that rule out US-based providers.

Pricing

Usage-based with three models: serverless (pay-per-token, with gpt-oss 120B at $0.15 per 1M input tokens and Qwen3 Coder 30B from $0.06), on-demand VMs (per-second billing, $3.59/hr for an H100 80GB, $1.69/hr for an L40S), and long-term contracts for reserved capacity. No minimum spend, and signing up needs no credit card. Prices as listed on lyceum.technology, August 2026, check their pricing page for current rates.

Who it is for

Teams that want EU-hosted inference and GPU compute without running their own hardware, from small AI startups shipping with serverless inference to infra teams orchestrating large training clusters. Lyceum is Berlin-based (Germany).

Frequently Asked Questions

Is Lyceum compatible with the OpenAI SDK?

Yes. Serverless inference exposes an OpenAI-compatible API, so you can point an existing OpenAI-style client at Lyceum, swap the base URL and API key, and stream completions without rewriting your integration.

Where does Lyceum host its GPUs?

In European data centres, with GDPR compliance and EU data residency. Inference and training both stay under EU jurisdiction, which is the main reason teams with sovereignty or compliance requirements pick Lyceum over US-based providers. The company itself is based in Berlin.

How does Lyceum pricing work?

Three models: pay-per-token serverless inference (gpt-oss 120B at $0.15 per 1M input tokens, Qwen3 Coder 30B from $0.06), per-second billing on GPU VMs ($3.59/hr for an H100 80GB, $1.69/hr for an L40S), and long-term contracts for reserved capacity. Prices as listed on lyceum.technology, August 2026.

Do I need a credit card or minimum commitment to start?

No. Signing up requires no credit card, billing is per second on compute and per token on inference, and there is no minimum spend.

Which models can I run on Lyceum?

The serverless API serves open-source models including gpt-oss 120B, Qwen3.5, Kimi K2.6, GLM-5.2, MiniMax M3, and DeepSeek V4, with more available in the dashboard. For models outside the catalogue, on-demand GPU VMs run any Docker container with full root access.

Can Lyceum handle large training jobs?

Yes. Beyond single VMs, Lyceum provisions clusters of 8 to 8,000 GPUs connected over InfiniBand for distributed training, with hardware ranging from L40S and A100 up to H100, H200, and the Blackwell B200 and B300.