Swiss AI Model Apertus Misunderstood, Says Expert

Critics slamming the Swiss AI model Apertus for lackluster chat performance are barking up the wrong tree. Since its high-profile launch on September 2, 2025, the model has faced a barrage of skepticism from users expecting a polished conversationalist. However, Philippe Cudré-Mauroux, a leading computer science professor at the University of Fribourg, asserts that these critics have fundamentally "misinterpreted" the project's DNA.

Apertus is not a Swiss ChatGPT designed to amuse the masses; it is a raw, industrial-grade engine built for scientists and enterprises. "The fact that several sites use the Apertus models to create a conversational agent has created confusion," Cudré-Mauroux states bluntly. While the public grapples with early, derivative chatbots that may hallucinate, the core technology remains a robust, generic Large Language Model (LLM) waiting to be fine-tuned. The expert emphasizes that distinguishing the core engine from these secondary, consumer-facing applications is critical. This is a tool for professionals, not a parlor trick.

In the brutal arena of global AI, numbers often dictate the narrative, and Apertus is fighting a strategic asymmetric war. The model's most powerful iteration boasts 70 billion parameters—a significant figure, yet it pales in comparison to the staggering 175 billion parameters of the previous generation GPT-3. But raw size isn't the only metric that matters.

According to Cudré-Mauroux, 70 billion parameters provide a "good base" capable of automating complex administrative tasks and spawning new, specialized AIs. The model, available on Hugging Face in both 8-billion and 70-billion versions, prioritizes accessibility over brute force. While it cannot currently match the sheer data volume of OpenAI or massive Chinese conglomerates—leading to lower accuracy in raw benchmarks—it offers a lean, adaptable foundation. The strategy is clear: provide a competent, malleable base for Swiss researchers rather than engaging in a futile arms race for the largest, most unwieldy model on the planet.

We are living in the "Far West" of international big tech, where opaque algorithms dictate digital reality. In stark contrast, Apertus stands as a fortress of ethics. It is one of the first 100% transparent models in existence, a claim that American and Chinese giants simply cannot make.

"Most American and Chinese AIs are very opaque," Cudré-Mauroux warns, highlighting the hidden biases and secret datasets that power competitors. Apertus flips the script by making its entire learning method public. This radical openness is not just a moral victory; it is a commercial necessity for local companies and researchers who demand absolute control over their data security. By rejecting the "black box" approach, the Federal Institutes of Technology (ETH Zurich and EPFL) and the CSCS are positioning Switzerland as the global safe haven for ethical AI development. Trust, in this volatile market, is the ultimate currency.

The clock is ticking on Swiss digital independence. While Apertus is a promising "first step," Cudré-Mauroux delivers a sobering ultimatum: funding must surge, or Switzerland risks irrelevance. Currently, the nation allocates significantly less capital to AI development than its industrialized neighbors, a gap that threatens to widen into a chasm.

"If Switzerland stays put, our self-determination is clearly threatened in the long term," the professor declares. We are not merely discussing software; we are discussing the ability of a nation to govern its own digital infrastructure without reliance on foreign powers. Apertus is not a finished product—it requires iteration, investment, and belief. Without a dramatic shift in financial support to refine these models and eliminate hallucinations, Switzerland may find itself a client state in the new digital world order. The technology is here, but the political will to fund it must follow immediately.