Blog · Published · Updated
PLLuM for business: licence or API?
Short answer
PLLuM makes sense when a company needs a Polish language model in a controlled environment and has a repeatable, high-volume task. The model files are free, but a GPU server with upkeep costs about €493 a month. At a case with 3,000 input tokens and 500 output tokens, the Claude Sonnet 5.5 API costs €0.0098, so your own server wins on price only from about 50,200 cases a month.
How we workDetails ↓
- 01Call30 minutes with an engineer, free
- 02PrototypeA working system on your process and data, free
- 03ImplementationTarget and fixed price in the contract; 50% at signing
- 04Acceptance50% once the agreed target is met
- model on your server or in the cloud in the EU
- quoted after the review
- same work in application with EU cloud model
- from €3,000
- task, data and equipment analysis
- €0
We belong to the OpenAI and Anthropic partner programmes and learn through their official courses. What this means ↓
Author: Syntalith · 3 min read
Statistics
Polish AI models in companies: statistics 2026
50,200
cases a month, from which a GPU server with upkeep at €493 comes out cheaper than the Claude Sonnet 5.5 API. Scenario.
17.8%
According to Statistics Poland (GUS, 2026), 17.8% of Polish companies use AI technologies.
8.4%
According to Eurostat (2025), 8.4% of Polish companies with at least 10 employees used AI.
95.9%
According to Poland's data protection authority UODO (2026), 95.9% of the organisations surveyed do not consider themselves ready to deploy AI in line with GDPR.
48%
According to the European Commission (2026), 48% of EU SMEs and mid-caps name a lack of expertise as a barrier to AI.
Source: European Commission, Flash Eurobarometer on AI (2026) ↗
Figures checked: 22 October 2026
First, the choice for the company
Control over documents
PLLuM in the company's environment
permissions, retention, deletion
A quick start, small traffic
a model through an API
the vendor's terms and the data flow
A steady, large volume
PLLuM and a comparison with the API
the cost of the server and its care
Hard reasoning, many languages
a stronger model through an API
quality and cost per case
PLLuM can handle one part of a process. Rules watch simple conditions, the model prepares an answer, and an employee approves the outcome. Such a split lets you measure the benefit without switching on the whole company at once.
The variant and the licence
PLLuM is a family of Polish language models developed for the public and private sectors. The PLLuM project site sorts the variants by licence, type and size. The instruct version follows instructions, chat is for conversation, and base is the base model.
| Variants in the 2512 series | Licence on the card | What it means for a company |
|---|---|---|
PLLuM-4B, PLLuM-12B | Apache 2.0 | commercial use on the licence's terms |
Llama-PLLuM-8B, Llama-PLLuM-70B | llama3.1 | the terms of the Llama 3.1 licence |
-nc- variants | non-commercial | research and private use |
The full name, series and licence are shown on the cards for PLLuM 12B Instruct, Llama-PLLuM 8B Instruct and Llama-PLLuM 70B Instruct. Record the card's address and the date you checked it in the project documentation. After a change of variant, check the licence again. All variants are in the PLLuM collection on Hugging Face.
What your own server costs, and what the API costs
- Hetzner GEX45 GPU server
- €214 a month net
- Server upkeep
- 8 h (€279)
- Your own model in total
- €493 a month
- One case on the Claude Sonnet 5.5 API
- $0.011 (€0.0098)
Server break-even
about 50,200 cases a month
Price of the GEX45 server with 24 GB of GPU memory from Hetzner's page on 22 October 2026, without the one-off €209 fee. €34.90 an hour is the EU average hourly labour cost for 2025 according to Eurostat. The API at Anthropic's rates, at the ECB rate of 1.1205 USD per euro on 22 October 2026. The upkeep time and the case size are the scenario's assumptions. Which PLLuM variant fits on this server, and at what quality, you check with a test.
A case: 3,000 input tokens and 500 output tokens. A GPU server with upkeep costs €493.
Źródło: Anthropic, API pricing (2026) ↗
PLLuM rarely wins on price alone with small traffic. It wins on the requirement that documents stay inside the company, or with a large, steady volume. We gathered the hardware details in the article Polish LLM hardware, costs and GDPR.
A quality test before launch
Prepare 30 to 50 anonymised examples from one process: typical cases, hard ones, incomplete ones and some where the correct answer is to ask a person for help. Compare a PLLuM variant with a second variant, with the API and with a simpler rules-based method.
- Correctness and completeness of the answer
- Agreement with the source document
- Behaviour when the documents give no basis
- The number of corrections an employee makes
- Response time and the cost of one case
The 4B variant may be enough for classification and short answers, and 12B handles longer context better but needs a larger environment. That is a hypothesis to check in a test on cases from your process, with the result recorded against the full version name.
Where the data sits and who is responsible
Running the model in the company limits sending documents to an outside service, but it does not settle GDPR compliance. Settle the purpose and the legal basis for processing, the roles, access, retention and deletion. With sensitive data, the article on GDPR and the DPA when deploying AI will help. We choose the variant and compare it with the API as part of private LLMs.
