A growing catalogue of open models, billed per token, with no monthly commitment. Or models we run on our own hardware, when your requirements reach the compute as well as the data.
Both run in Sweden, on the same OpenAI-compatible API. Pick the one that matches your constraints.
A catalogue of open models you reach with one API key. Pay for the tokens you send and receive. Nothing to commit to, nothing to plan capacity for.
Models we run ourselves, on hardware Aixia owns and administers in Aixia datacenters. The answer when the requirements cover where the compute runs, not only where the data rests.
What holds true whichever way you buy.
Models and hardware run in Sweden, under GDPR and ISO 27001, with no dependency on a non-European cloud. Nothing leaves the country on either track.
New models reach the catalogue as they prove themselves. Moving to one is a change of model ID, not a change of architecture.
Support for model testing, validation, optimization, and deployment from Aixia's AI experts.
Start on tokens with nothing committed. Move to a subscription when the workload settles. Neither direction is a migration.
Published per-model rates and per-key metering. You can see what a workload costs before you run it, and what it cost after.
Everything we serve is open weight, behind an OpenAI-compatible API. Nothing here is a proprietary endpoint you cannot leave.
Rates in SEK per million tokens, excl. VAT. You are billed on the tokens you actually send and receive, per API key.
| Model | Input / 1M (SEK) | Output / 1M (SEK) | Good at |
|---|---|---|---|
Kimi K2.6kimi-k2.6-swe |
13.86 | 55.45 | Reasoning, coding, long context. The one to reach for first. |
Gemma 4 26Bgemma-4-26b-swe |
1.44 | 5.55 | Fast and cheap. Built for high volume and short turnarounds. |
Qwen3.8 27Bqwen3.8-27b-swe |
8.32 | 33.27 | General purpose. A capable default for mixed workloads. |
Qwen3 Embedding 8BQwen3-Embedding-8B-swe |
1.11 | 1.11 | Multilingual semantic search and RAG retrieval. |
Qwen3 Reranker 4BQwen3-Reranker-4B-swe |
0.55 | 0.00 | Reranking retrieved passages before they reach the model. |
Catalogue current as of August 2026. It is extended as new models prove themselves. Reranking is charged on input only. Models and hardware are hosted in Sweden. Catalogue compute is operated by a Swedish partner under a data processing agreement. If your requirements rule out a processor entirely, that is what the in-house subscription is for.
Models we run on hardware Aixia owns and administers. A dedicated VPS for agents, retrieval and preprocessing is included in every tier. All prices excl. VAT.
What you get on each track, and who each one is for.
Per-model rates
No monthly fee, no included allowance
Purpose: Scaling usage without committing capacity
Best for: Teams whose volume moves week to week, and anyone who wants the newest models without a migration
1 500 SEK / month
1M tokens included (max 10M)
Purpose: Proof-of-Concept, innovation, and testing on in-house models
Best for: Organizations exploring AI capabilities or validating use-cases
3 500 SEK / month
10M tokens included (max 50M)
Purpose: Production workloads and multiple parallel use-cases
Best for: Companies running LLMs in production with higher availability needs
7 500 SEK / month
50M tokens included (unlimited)
Purpose: Business-critical, high-scale operations
Best for: Organizations where AI is mission-critical and performance cannot be compromised
Take a key and start on the catalogue today. Talk to us when the workload needs to sit on hardware we own.