AIMAX^®
- AI providers and AI models

AIMAX®’s model-agnostic approach offers you the greatest possible flexibility and, consequently, maximum autonomy. Whether you’re looking for maximum control thanks to local AI, the ideal balance between data protection and the performance of large open models from your trusted AI provider, or the frontier models primarily offered by US hyperscalers – with all their advantages and disadvantages.

Even any combination in between, with a choice of provider and model tailored to the specific task, performance requirements and data protection needs, is possible with AIMAX®. You also determine, at a granular level, which users and/or roles are granted access to which AI models and whether, where applicable, a request and manual confirmation should be required beforehand.

This goes as far as, for example, allowing a powerful Frontier model to be explicitly integrated into a workflow specifically for the creation of marketing images; whilst employees are granted authorisation to use this workflow, they are not granted the rights to ‘misuse’ the Frontier model for any purposes of their own.

To meet the most stringent data protection requirements, right through to purely local operation, open AI models such as Qwen, Gemma, Whisper, Muse Glimmer, etc. can be run on AIMAX® in a wide variety of configurations. We, or your AIMAX® partner, will work with you to select the model that best suits your needs.

AIMAX® is capable of running these models in parallel across multiple – even different – hardware systems, including load balancing and failover. This also allows you to scale your inference hardware resources gradually as required – for example, in the event of increased usage due to greater automation of your processes.

There may be a variety of reasons for using open models with an AI inference provider or AI data centre of your choice. If, for example, you wish to deliberately avoid purchasing expensive inference hardware or to delay doing so, using a German or European AI provider is an alternative that complies with data protection regulations and is also financially very attractive. 

Furthermore, some AI data centres also offer models whose performance – particularly in terms of VRAM – quickly outstrips that of ‘normal’ local hardware. Examples include GLM 5.3, Kimi K3, DeepSeek V4.1 Flash, etc. In terms of performance, these models are already close to the major frontier models or are only a few weeks behind in this respect.

This approach combines both high data protection standards – guaranteed by a German or European provider with no possibility of interference under the US CLOUD Act – and the high performance of the models.

Using these large open models on high-performance data centre hardware can certainly serve as the perfect complement to on-premises AI. For organisations with a large workforce, for example, it may make sense to rely on an AI data centre for direct interaction, thereby ensuring short response times with a comparatively very powerful model, whilst automated processes run in the background – albeit somewhat more slowly but still reliably – on local AI.

This option represents a potential compliance-related data protection compromise and reduces the usage-based token costs incurred with the AI provider. That said, to be honest, you really do need to do the maths carefully to work out when having your own inference hardware is actually worth it, based purely on a comparison of token costs.

Below is a list of potential German and European AI providers that you can use in AIMAX® (this list is not exhaustive).

In addition to the providers already pre-configured in AIMAX®, you generally also have the option of adding your own providers or inference endpoints, provided they offer an OpenAI-, Anthropic- or Ollama-compatible interface.

Frontier models from OpenAI, Google and Anthropic are at the forefront of AI development. Currently, providers such as Mistral, DeepSeek, Moonshot AI and Alibaba are in the second tier. It is hardly worth listing the models, as this list is already out of date by the time of writing.

If you require maximum performance, the cloud-based models offered by these providers are the best choice. However, caution is advised here, as – with the exception of Mistral – these are US or Chinese providers. The Microsoft Azure Foundation occupies a special position here, offering selected third-party frontier models – for example, from OpenAI – on the basis of partnerships, some of which are hosted in European data centres.

AIMAX® also offers you suitable support for this class of models. The providers OpenAI, Microsoft Azure Foundation, Anthropic and Mistral are already pre-configured in AIMAX®. Via a settings option, you can specify that manual consent must be obtained before these providers are used and that staff must be made aware of this accordingly. Furthermore, you have the option to restrict data processing to the EU or ‘EU-flex’. EU-flex allows the use of providers outside the EU. However, as soon as the AI attempts to access your company’s data, data processing is automatically – and then irreversibly – restricted to EU-based providers.

Below is a list of potential Frontier model providers that you can use in AIMAX® (this list is not exhaustive).

Here too, the following applies: in addition to the providers already pre-configured in AIMAX®, you generally also have the option of adding your own providers or inference endpoints, provided they offer an OpenAI-, Anthropic- or Ollama-compatible interface.