Switzerland's Giotto announced its Solo and Alto reasoning models on October 27. According to the company, Solo is designed for one GPU, while Alto runs on eight B300 GPUs. A free anonymous chat is also offered. These hardware descriptions are supplier statements, rather than a performance comparison conducted by SIExpo.
Different ways to use the system
The separate Offering page presents three routes: a European cloud environment with managed GPUs and updates, deployment on customers' compatible GPUs, and ready-to-use workstations or servers with software and support. Downloading software for private infrastructure should not automatically be described as an unrestricted license or publicly released weights with unlimited usage rights.
Our editorial reading is that choosing an execution environment changes responsibilities. In a managed service, some daily operations remain with the provider. With a private installation, the organization must establish who updates the system, monitors failures, and maintains recovery. Choosing a country does not settle those questions or define the complete operational arrangement.
In our example, an accounting team compares two ways to process documents without real personal data. One uses a managed environment and the other a customer-controlled server. We would evaluate identical documents and extraction criteria, as well as setup and correction time. This is a proposed method, rather than a description of an actual Giotto customer or a trial we conducted.
Reading the sovereignty claim carefully
The privacy policy addresses categories of recipients and allows international transfers subject to the stated safeguards. It also specifies different retention periods according to processing purpose. General positioning as European AI therefore does not establish that every category of data in every service remains indefinitely within one country, under one identical retention rule.
We make no legal determination about a specific deployment and have not reviewed a private customer contract. The practical question is how the selected arrangement applies to actual data flows. For example, a document, model input, operational log, and payment information can serve different purposes. Describing them separately is more useful than covering all of them with a single general label.
Small and large configurations are different services
A single accelerator and a multi-accelerator configuration raise different workload questions. Our conclusion is to compare accepted tasks under the same response-time requirement. A larger resource may support more concurrent work, but its presence alone does not determine the quality of an individual answer or the cost of operating the service.
For a pilot, we would record the exact version, settings, input sizes, and cases in which the model should acknowledge insufficient evidence. We would also include a document changed after initial indexing. The evaluation could then distinguish persuasive language from an up-to-date result connected to the correct source version, with correction work included in the assessment.
The published comparison tables belong to the manufacturer; we do not turn them into independently established universal superiority. The announcement broadens European choices of models and execution environments. Suitability for an individual team will depend on verifiable task results, the specific license, support, and agreed data conditions. Those elements need to be considered together before a deployment choice can be evaluated meaningfully.
*AI illustration: a conceptual editorial image, not a documentary photograph or an actual product screen.*
*Editorial verification: 2026-10-28T17:14:21.370659+00:00.*
