Last updated: June 14, 2026
AffordableAI gives you access to open-weight AI language models through an OpenAI-compatible API (and an Anthropic-compatible one). You get up to 10 concurrent requests, reasonable usage, and email support when we can. The model might change — we might upgrade to a newer open-weight release or switch model families. We'll tell you when that happens. The API stays compatible.
Use real information when you sign up. Keep your API keys safe. Don't share them. Anything that happens under your account is your responsibility. We'll suspend accounts that break these terms.
It's €20 per month, VAT included where applicable. Mollie B.V. handles the actual payment processing. Cancel whenever — your access keeps working until the current billing period ends. We don't refund partial months. If a payment fails, there's a grace period before we cut off access.
Don't do illegal things under Dutch or EU law. That includes generating malware, phishing content, spam, CSAM, hate speech — the obvious stuff. Don't try to bypass rate limits or fair-use controls. Don't resell API access unless you've got a commercial agreement with us. Don't use the service to train a competing model. And don't run it in a way that degrades things for other people on the same hardware.
This is a flat-fee service with no usage caps. That only works if we set some guardrails. Otherwise a handful of heavy users would eat the whole GPU and everyone else gets a degraded experience.
The limits: 10 concurrent requests per user (extra requests queue up behind them). Token generation speed is shared across everyone on a given server. Under normal load you'll see 75-165 tok/s — faster than ChatGPT. If you're running continuous maximum-concurrency 24/7, doing automated scraping, or doing something that looks like a DoS attack, we might rate-limit you or review your account.
These limits affect almost nobody. Under 1% of users ever hit them. If you've got a legitimate use case that goes beyond fair use, email us. We can set up dedicated capacity.
We don't own your inputs. Prompts, code, documents, whatever you send to the API — it's yours. We claim no ownership, no license, nothing.
We don't own your outputs. Completions, generated code, whatever the model sends back — also yours. We don't claim any rights to it.
The model weights aren't ours either. We run open-weight models (currently MIT-licensed). We didn't make them, we just operate them.
There's a reason we can be this definitive: we don't store any of it. No prompts. No completions. No training. It passes through GPU memory and it's gone. There's nothing to claim ownership over.
We don't log, store, retain, or archive: prompts, completions, chat history, conversation content, files, code, or documents submitted to the model.
Prompts and completions live in GPU volatile memory during the request and get discarded immediately after. We don't train on your data. We don't use your data to improve the service. The Privacy Policy and DPA have the full treatment if you want the details.
We aim for high availability but there's no SLA on the self-service tier. Enterprise customers can negotiate uptime commitments. We try to give notice before planned maintenance.
The service runs on rented GPUs in Finland. If the provider has an outage, we're down until the instance comes back or we spin up a replacement. We monitor things and respond to incidents quickly.
To the extent the law allows, we're not liable for service downtime or data loss (though there's not much to lose — we don't store your content). Not liable for model outputs that turn out to be wrong, offensive, or otherwise problematic. Not liable for indirect, incidental, or consequential damages, or for lost revenue, profit, or business.
Our total liability for any claim is capped at whatever you paid us in the 12 months before the claim.
You can cancel anytime. We can suspend or terminate your account if you violate these terms, with notice where reasonable. After termination, account data is deleted within 30 days. Billing records stay as long as Dutch law requires.
We'll update these when needed. Material changes get an email. Keep using the service after a change means you're okay with it.
Dutch law. Disputes go to courts in Amsterdam.
Questions: hi@affordableai.eu