Model comparison · 2026-09-08
How to Use GreenPT Models Through LLMTR
A developer's view of the GreenPT rows in the LLMTR catalog: the measured behavioural differences and the three cost surprises that affect your requests.
What GreenPT is and how it appears in the catalog
GreenPT is an Utrecht-based platform that states it runs its models inside the European Union. The LLMTR catalog carries its chat, embedding and reranking rows, all addressed with the greenpt prefix.
The provider runs no hardware of its own; it sits in front of contracted data centres and routes each request to a suitable one. That helps on price and model variety, and it lengthens the chain for a data-protection assessment.
The -eu suffix on model identifiers
Some GreenPT identifiers end in -eu. The reason is not cosmetic: we also serve the same model through another provider in the catalog, and a model identifier sent without a provider prefix also resolves.
If two rows shared one identifier, an unprefixed request would reach whichever came first — and since the prices differ, your bill could change without you noticing. Use the identifier exactly as the catalog spells it; searching for the unsuffixed name still finds the row.
Three behaviours that drive cost
Three things were measured on this provider that enlarge a bill unexpectedly, and all three are worth knowing before you choose a model.
None of them is a fault; they are the provider's design choices. But none appears in a price table either, so they are hard to notice before the first invoice.
- 26 of the 34 chat rows reason whether you ask them to or not, and those hidden tokens bill as output.
- Some rows run under a system instruction injected server-side, billed to you as input on every request; in measurement a one-word request was billed as 8,082 input tokens on one row.
- On the instruction-free equivalent of that row, the same request cost 72 tokens; if you write your own system prompt, choose the instruction-free row.
- A narrow output budget can leave the answer completely empty: some rows spent an entire 256-token budget on reasoning.
Rate limit and practical notes
The provider enforces 600 requests per 300 seconds; the sustainable rate is roughly two requests per second. In measurement three concurrent requests were enough to hit the limit, so queue on your side for high-concurrency workloads.
On reranking there are two behavioural differences: the provider makes the top-n field mandatory, and LLMTR fills it in for you from the document count; and the truncation parameter is not accepted, because the provider fails the whole request rather than shortening a long document.
Frequently asked questions
Do GreenPT rows need a separate key?
No. You call them with your existing LLMTR key by changing the model identifier. You do not need an account with the provider.
Why do some identifiers end in -eu?
Because we also serve the same model through another provider in the catalog. Since unprefixed requests also resolve, two rows sharing one identifier could change your bill without you noticing.
Which rows spend no reasoning tokens?
Eight of the 34 chat rows, and they are named on the documentation page. Prefer them for batch work where output cost has to be predictable.