
Open Code + Lyceum Technology
If you are looking for a European alternative for Claude Code or Codex, you might have seen Mistral Vibe already. I found another promising option now, that I like a lot.
lyceum.technology offers serverless compute, which basically runs OpenAI compatible open LLMs incl. some very strong ones from China.
The experience together with Open Code, a coding agent which you can use from console as well from a web interface is super smooth.
What I like a lot about it is ...
- it is hosted in the EU
- there is no fixed monthly fee, instead you pay only as much as you need
- you can switch between different models to save money
- tokens are cheaper in general compared to the US providers
- LLMs en par with OpenAI/Anthropic models
If you want to try it:
- Register at lyceum.technology and add 10€ to your account
- Generate an API key: https://dashboard.lyceum.technology/api-keys
- Copy it or keep the page open, we will use it later
- Get Open Code here for free: opencode.ai
- Add lyceum to your configuration (feel free to copy the following text into
~/.config/opencode/opencode.jsoncto use GLM-5.2 -> imho comparable to GPT-5.5, see https://artificialanalysis.ai/leaderboards/models )
​
{
"$schema": "https://opencode.ai/config.json",
"autoupdate": true,
"permission": {
"edit": "ask",
"bash": "ask"
},
"model": "lyceum/GLM-5.2",
"provider": {
"lyceum": {
"npm": "@ai-sdk/openai-compatible",
"name": "Lyceum Technology",
"options": {
"baseURL": "https://api.lyceum.technology/api/v2/external/serverless",
"timeout": 300000,
"maxRetries": 3,
"retryDelay": 1000,
"setCacheKey": true
},
"models": {
"z-ai/glm-5.2": {
"name": "lyceum/GLM-5.2"
}
}
}
}
}
- Start
OPENCODE_ENABLE_EXA=1 opencodefrom a terminal - In the Open Code interface enter
/providerand search for "lyceum" - Select
lyceum/GLM-5.2and if asked, enter the API key you received above.
Now, you can start using Open Code with an LLM hosted in Europe.
I am not affiliated or in any relation with Lyceum Technologies other than being a happy customer! 😄
A little hint: part of what makes Claude Code successful is, that they are adding some instructions to every prompt. Open Code does not do this. So my advice is, create a global AGENTS.md, which describes how to work properly. You can ask Open Code to create it for you. My advice would be to ask it to plan things together with you, parallelize where possible on execution, exchange how you will review the result later and always to verify its result. This will increase overall quality. I don't want to go to deep into this, but there are plenty of people here on Reddit who probably can share their experience.