Solheim.ai has launched the Virtual Private LLM service — a dedicated LLM instance on servers in Europe with a fixed monthly subscription, no token meter, and no limits on request volume. Pricing depends on the number of parallel instances and the size of the context window, with a starting price of €15/month. The server infrastructure is located in European data centers, and requests do not leave the EU jurisdiction.
What happened
Solheim.ai has released the Virtual Private LLM service, which provides a dedicated LLM instance for a fixed subscription. Pricing is based on two parameters: the number of parallel instances and the size of the context window. The basic plan is €15/month for one instance with a 64k context. Two models are available at launch: Qwen3.6-35B-A3B with MoE architecture (35B parameters, 3B active per token, Apache 2.0 license) and DeepSeek V4 Flash with MIT license and 256k context. The API is fully compatible with the OpenAI SDK, and connecting to Cline, Roo Code, and VS Code is done by replacing two environment variables.
Context
The LLM inference market has long been based on a token model: payment depends on the number of tokens consumed, and a rolling usage window leads to unpredictable costs and sudden blocks in the middle of a session — for example, Claude Code may stop due to exceeding limits. Solheim offers an alternative based on the principle of rent the machine, not the tokens — you rent computing resources rather than paying for each token. The MoE architecture of Qwen3.6-35B-A3B theoretically provides efficient inference by activating about 9% of parameters per token, which justifies the economics of a fixed subscription with 24/7 availability.
Why this matters for the industry
The service addresses the data sovereignty issue in inference: requests remain within the EU jurisdiction, meeting GDPR and EU AI Act requirements without the involvement of infrastructure subject to the US CLOUD Act. The VPS for LLMs pattern could become a standard for EU-first AI products. If the rent the machine concept spreads, it will put pressure on the pricing of US hyperscalers and could shift the industry from token-based to instance-based billing, especially for open-weight models.
Why this matters for users
For teams that need a predictable LLM backend for coding agents, automation, or prototypes, Solheim offers an OpenAI-compatible API from €15/month with European compliance. Switching from another provider takes minutes — just replace the API key and base_url. The open-weight models Qwen3.6 and DeepSeek V4 Flash are inferior to Claude and GPT on complex tasks, but they work well enough for typical dev tasks (diffs, tests, refactoring, tool-calling). Predictable pricing eliminates billing surprises that arise with token-based models.
What is still unknown / limitations
Solheim does not publish performance benchmarks: tokens per second, latency, time-to-first-token, and SLA uptime remain unknown. There is no data on server hardware specifications — GPU class, number of cards, and inference stack configuration (vLLM, TensorRT-LLM, or other) are not disclosed. Without these parameters, it is impossible to assess under what loads the fixed price remains profitable for the provider and what actual inference quality users receive.
Sources
- Solheim — your own EU-hosted LLM instance
- Building Solheim: a VPS for LLMs, hosted entirely in the EU — Indie Hackers
- Show HN: Virtual Private LLM, fixed fee with no usage or token limits — Hacker News
Author
Look at AI, editorial team