Terms of use

Last updated: September 24, 2026

The service

Benchmark Heaven is a free website operated by productivity-boost.com Betriebs UG (haftungsbeschränkt) & Co. KG (see Impressum). It collects published benchmark results and prices for AI models and shows a modeled cost per task.

No guarantee of accuracy

Every number comes from a named source with a date, and modeled costs are estimates. Sources change and can be wrong. The information is provided as is, without warranty, and is not advice for any purchase or business decision. Check the original source before relying on a value. We are liable without limitation for intent and gross negligence, and under the statutory provisions for injury to life, body or health; otherwise liability for a free service is excluded as far as the law permits.

Accounts

An account is optional and free. It stores your saved presets and settings. You can delete it at any time on the Account page. Please do not misuse the service, for example by automated mass requests that impair it; the public API is the intended way to use the data programmatically.

Priority evaluation requests

Priority evaluation is an optional paid service for earlier scheduling. It does not change the benchmark method, task set, score or rank. We decide which models we can evaluate and when, and we may evaluate a model on our own schedule if it is of public interest. The regular community benchmark remains free.

The fee is USD 49 per selected benchmark for API-served models or open models up to about 9B parameters, or USD 99 per selected benchmark for larger open models that we run on our GPUs. Selecting both JevBench and ImageJevBench incurs the selected fee twice. Applicable taxes are calculated at checkout. Payment is completed through Stripe Checkout; Stripe sends the payment receipt to the email address provided at checkout.

Every submission is reviewed before evaluation. We may refuse a submission that is unsafe or cannot be evaluated fairly, and we will issue a full refund. After a submission passes code review, we provide its results within 48 hours. If we miss that deadline, we automatically issue a full refund. The 48-hour period starts when we tell you the code review has passed.

For a public request, you authorize us to publish the resulting aggregate leaderboard row, marked “priority run”. A private request produces a report for your team and is not published without your consent. We may still evaluate the model later on our own schedule if it becomes a matter of public interest. You must not submit API keys, passwords or access tokens through the request form. Any later credential handover must use the encrypted process we provide.

Privacy

How we handle personal data is described in the privacy policy.

Law

German law applies. Mandatory consumer protection rules of your country of residence remain unaffected.