Swiss AI Initiative adds image and audio capabilities to Apertus 1.5

The fully open language model was trained on the Alps supercomputer and is available under an Apache 2.0 license for research and commercial use

Apertus 1.5 adds image and audio processing to the fully open language model developed through the Swiss AI Initiative

EPFL, ETH Zurich and the Swiss National Supercomputing Centre have released Apertus 1.5, adding the ability to process images and audio alongside text to the Swiss AI Initiative’s large language model.

The update expands a model designed to provide an open alternative to proprietary AI systems. Apertus gives researchers, public institutions and companies access to its training data, code and development process, allowing them to inspect and adapt the technology.

Apertus 1.5 was trained on the Alps supercomputer at the Swiss National Supercomputing Centre, or CSCS, in Lugano. The project is led by Imanol Schlag, a research scientist at ETH Zurich, and EPFL professors Martin Jaggi and Antoine Bosselut.

“Rather than introducing an entirely new model, Apertus 1.5 incorporates feedback from the first release and adds several new capabilities,” Jaggi explains.

Beyond its new multimodal functions, the team says the model offers improved reasoning, instruction following and support for using tools.

An open alternative to commercial AI models

Apertus is positioned as part of a longer-term attempt to establish sovereign AI infrastructure, giving organizations an option that can be examined and operated without relying entirely on proprietary commercial models.

“Apertus is not about competing with frontier models from private companies, but about providing trustworthy, transparent AI that organizations big and small can build upon with confidence,” Schlag says. “With every release, this foundation becomes more capable.”

The team plans to release further versions over the coming years, with expanded capabilities intended for research, education, public administration and industry. The announcement does not identify specific education deployments for Apertus 1.5.

The model is released under the Apache 2.0 open-source license, which permits research and commercial use. The full suite is available through Hugging Face.

Apertus Mini has also recently been introduced alongside the main model. The suite contains 16 compact models intended to demonstrate how distillation and quantization can support deployment on more limited hardware.

Existing deployments put openness into practice

Earlier versions of Apertus are already being used by public institutions, researchers and media organizations in Switzerland.

The Canton of Ticino uses the model for an internal AI translation service, allowing sensitive government documents to be processed without using a commercial provider. At EPFL, researchers have used Apertus as one of the foundation models for MeditronFO, a fully open framework for building medical large language models.

Basel-based news outlet Bajour runs Apertus locally to help prepare political coverage for its Basel Briefing newsletter. It has also developed a search and analysis tool that allows journalists to examine cantonal parliament transcripts, track politicians’ positions and compare statements made over time.

“As journalists, we value transparency and openness,” says Bajour editor Samuel Hufschmid. “With Apertus, we know what we're building on. Its open-source nature and transparency make it a natural fit for our newsroom.”

Bajour’s tools are also being made available to other news organizations through the We.Publish publishing ecosystem.