When I started using LLM based AI agents I first tried OpenCode for the following reasons:
- It's open source, which I generally prefer
- It works with plenty of different LLM providers
I initially used it with the default OpenCode Zen provider, but quickly hit limits of the free plan. I then got myself a Claude Pro subscription and used it with that. But at some point Anthropic disallowed using their subscriptions with tools other than Claude Code.1
After trying to use an Anthropic API key for some time, I noticed how much more expensive it is and switched to using Claude Code and kept using the subscription.
But this whole thing made me realize how dependent I am on a single company which almost forces me to use their proprietary tool. On top of that I'm sending quite a lot of data to them through my use of their services.
All that led me to research which providers of LLMs exist in Switzerland. https://mydata.ch/schweizer-ki-anbieter gives a nice overview of some existing providers. Infomaniak stood out because they have an affordable pay by use plan and have some decent open weight models available.
Prerequisites
You will need the following:
- An Infomaniak account with AI Services enabled (1 million free credits are included)
- Your AI
product_idfrom the Infomaniak Manager - An API token scoped to AI Services
- OpenCode installed
Configuring the Provider
Add the following to your ~/.config/opencode/opencode.json:
{
"$schema": "https://opencode.ai/config.json",
"provider": {
"infomaniak": {
"npm": "@ai-sdk/openai-compatible",
"name": "Infomaniak AI Tools",
"options": {
"baseURL": "https://api.infomaniak.com/2/ai/{product_id}/openai/v1"
},
"models": {
"moonshotai/Kimi-K2.6": {
"name": "Kimi K2.6",
"limit": { "context": 256000, "output": 32768 }
}
}
}
}
}
Make sure to replace {product_id} with the numeric ID of your AI Services product.
Run /connect in the OpenCode TUI
and select our custom provider.
In the next screen paste your API token which you can create with the Infomaniak Manager as well.
Using It
My first impression was that it feels faster than using Claude Code with the Opus models. According to benchmarks Kimi K2.6 should be comparable to the Opus 4.x models.2
I had one occasion where it seemed to block for around a minute on a very simple task. Maybe the server was overloaded there? Other than that it felt faster than using Opus 4.8 with my Claude Pro subscription.
What I'm missing from Claude are:
- That it uses interactive forms heavily during plan mode to ask me questions where I can just select one of the suggested answers or provide my own.
- Auto-mode, permission fatigue is real and together with nono.sh sandboxing auto mode feels like a very good compromise to me.3
I guess there are plugins for OpenCode to achieve a similar feature set as I'm used to from Claude. https://github.com/the-perfect-developer/the-perfect-opencode/tree/main looks promising for example.
Also I'm not sure what differences in the experience are due to the different LLM and which due to the changed harness. To more objectively compare the models I should probably use both with OpenCode.
Costs
Comparing the costs per token is quite straightforward (USD converted at CHF 0.81 per US dollar, as of September 2026):
| Provider | Infomaniak Kimi K2.64 | Anthropic Opus 4.85 |
|---|---|---|
| Input | CHF 0.60 / 1M | $5 (CHF 4.05) / 1M |
| Output | CHF 3.00 / 1M | $25 (CHF 20.25) / 1M |
So Anthropic is almost 7x as expensive compared to Infomaniak.
For August 2026 I used my 1 million free credits in two days and the rest of my usage came to CHF 16.10. But for September, where I started using it more frequently and for bigger tasks, I already reached CHF 22.00 on the 10th, probably ending up using around CHF 60. When using the Anthropic API this would result in approximately CHF 400, assuming roughly the same token usage.
But with the Claude Pro plan it looks different: compared to the fixed price of the Claude Pro subscription of CHF 14 per month, Infomaniak costs me around 4x as much.
Summary
I love being able to use a Swiss hosted open weight LLM for the following reasons:
- Digital sovereignty: I feel less dependent on a single American company
- No vendor lock-in: I can freely choose which agentic CLI to use
- Supporting a domestic company: I like to support local businesses
While I miss some of the power Claude Code gives me with their frontier models, I think in the long-term I'll benefit more from using open weight models with open tools. The downside is of course that it is way more expensive than using the Claude Pro subscription.
-
See discussion on HN from that time, the OpenCode Anthropic provider docs
There are plugins that allow you to use your Claude Pro/Max models with OpenCode. Anthropic explicitly prohibits this.
and the Anthropic ToS
You may not access or use, or help another person to access or use, our Services in the following ways: ... 7. Except when you are accessing our Services via an Anthropic API Key or where we otherwise explicitly permit it, to access the Services through automated or non-human means, whether through a bot, script, or otherwise.
-
See https://benchlm.ai/compare/claude-opus-4-6-vs-kimi-2-6, https://codersera.com/blog/kimi-k2-6-vs-claude-opus-4-7/, https://codingfleet.com/blog/claude-opus-4-8-vs-kimi-k2-6/ ↩
-
See my previous blog post Safe Ways to Use AI Agents on the topic of sandboxing ↩
-
See the Infomaniak AI Tools pricing ↩
-
See the Claude Platform pricing ↩