This is a brief list of places to get access to Large Language Models (LLMs), in particular via API, for the physics/research/academic community.

General Providers

They provide broad access, usually via a SSO provider such as Helmholtz AAI. Members of of specific communities may be provided with better access (more/commercial models, higher rate limits).

GWDG (Göttingen): SAIA

The Gesellschaft für wissenschaftliche Datenverarbeitung mbH Göttingen (GWDG) facilitates access to a large amount of LLM models, both commercial (i.e. ChatGPT, ...) and self-hosted (i.e. GPT-OSS, Qwen, ...).
The self-hosted models are, generally speaking, available to any user with an institutional account, via Academic Cloud SSO.

The models can be access via a website or via API through a service called SAIA.
For the latter one needs to apply with the KISSKI LLM Service where some details are asked for and access is typically granted within at most a few days.

Rate limit: indicated in the response header, see documentation.

AI@JSC (Jülich): Blablador

The Jülich Supercomputing Centre (JSC) offers access to a fair selection of LLM models.

They can be accessed via a website or via API, which requires only a brief setup.

Rate limit (see documentation):

• degraded: 5-15/h (when usage caused issues)
• external: 15-30/h (anybody)
• internal: 50-100/h (all users from FZJ)

Exclusive Provides

They provide access to select communities.

KIconnect.nrw (Aachen): RWTHgpt, ...?

This service provides access to a few LLMs for participating institutions in North Rhine-Westphalia (NRW), e.g. RWTH Aachen, ...

It can be accessed via a website and via API, which can be easily created.

Rate limit (see model selection menu on website):

• RWTHgpt
  • selfhosted (GPT-OSS, Mistral small): unlimited
  • commercial (GPT-5.x): 10-50/h 25-150/day

Available models and limits may vary depending on your home institution.