Three questions, one clear answer: which cloud AI is permitted for your industry, where your data is processed and how strong the models really are on that route.
Humanyx guides financial service providers as a partner into an AI-enabled operating model. This overview shows all common routes, including those we do not implement ourselves.
Industry rules and supervision feed directly into the traffic light.
IndustryWealth managers and financial service providers
Health data appears as soon as you choose insurance brokers.
Your internal policy, not the law. We show both.
Our recommendation: client data only in Switzerland. Processing abroad is only permitted with a risk assessment, a contract and protection of the client data, and a breach of professional secrecy is a criminal offence (Art. 69 FinIA).
Top score today: 58 with Claude Opus 5.5, available with EU processing (traffic light yellow)
PreviewIn the preview at Phoeniqs, DeepSeek V4.1 Flash (45), GLM 5.3 (45), Kimi K3 (44) are running.
The number is the Artificial Analysis Intelligence Index: an independent benchmark with ten demanding task sets covering agentic tasks, office work, knowledge, coding and science. Scale 0 to 100, the best model scores 58 today. As of 23 October 2026, version 4.3.2. "ca." = estimated.
Our recommendation: client data only in Switzerland. Processing abroad is only permitted with a risk assessment, a contract and protection of the client data (see the Swiss Bankers Association's Cloud Guidelines), and a breach of professional secrecy is a criminal offence (Art. 69 FinIA).
Sources: Microsoft Learn: Deployment types in Microsoft Foundry Models (Global, Data Zone, Standard, Provisioned)Microsoft Learn: Data residency for Microsoft 365 CopilotMicrosoft Learn: «Store Microsoft 365 customer data in-country» settingMicrosoft Learn: Flex routing (EU and EFTA)FINMA: Circular 2018/3 OutsourcingSwiss Bankers Association: Cloud Guidelines (3rd edition, November 2025)Fedlex: Financial Institutions Ordinance, Art. 17 (outsourcing)Fedlex: Financial Institutions Act, Art. 69 (professional secrecy)
Each dot is an option, placed by data location and by the performance of the best model on that route. The colour shows the traffic light for your selection above.
Artificial Analysis Intelligence Index. Independent benchmark by Artificial Analysis. All models solve the same ten demanding task sets: agentic tasks and real office work (30%), general knowledge, long documents and hallucinations (30%), coding (20%) and scientific reasoning (20%). Scale 0 to 100, higher is better. The tests are deliberately very hard: the best model scores 58 today.
Version 4.3.2, values as of 23 October 2026. "ca." = estimated by Artificial Analysis, the independent measurement is still pending.
At Phoeniqs with 42. Azure Switzerland only offers older GPT models: GPT-5.1 (ca. 25, Provisioned only) and GPT-4o (ca. 8). The newest GPT models only run with EU processing.
Phoeniqs is testing Kimi K3 (44) until 16 November 2026 without an SLA, and production use is not planned according to the documentation. DeepSeek V4.1 Flash and GLM 5.3 (45 each) are meant to go into production.
Claude Opus 5.5 via AWS Bedrock or Google Vertex AI with EU processing reaches 58, the highest value in the index. Traffic light Yellow: feasible with a risk assessment and pseudonymisation.
The traffic light applies to your selection above. Filter by colour and open an option for the data path, models and checklist.
Green: proven and recommended
Works with client data: the setup is proven, contract and governance are in place. This is exactly how you should do it.
Yellow: unclear, depends on the setup
Depends on the contract, configuration and industry: feasible with a documented risk assessment, not without it.
Red: not possible
Not defensible for client data in our risk assessment: no contract, no control or no documented safeguards. Often still the best choice for research, marketing and anonymised content.
Microsoft, OpenAI, AWS and Google are US corporations and are subject to the US CLOUD Act, even with data centres in Switzerland or the EU. This residual risk belongs in every risk assessment.
Performance: Artificial Analysis Intelligence Index, version 4.3.2, values as of 23 October 2026, scale 0 to 100. "ca." = estimated by Artificial Analysis. 128,000 tokens correspond to roughly 190 A4 pages. The traffic light is our risk assessment and does not replace legal advice. Data as of: 23 October 2026.
Not listed: Meta Muse Wonder (Index 52) and Grok 4.7 by xAI. Both are among the strongest models but currently have no deployment route with processing in Switzerland or the EU.
This overview does not replace legal advice. Model availability and regions change constantly; October 2026 edition. Verify details with the provider before use.
Sources: FDPIC: Cross-border transfer of personal dataMicrosoft Learn: Region availability for Foundry Models sold by AzureMicrosoft Learn: Deployment types in Microsoft Foundry Models (Global, Data Zone, Standard, Provisioned)Microsoft Learn: Data residency for Microsoft 365 CopilotMicrosoft Learn: «Store Microsoft 365 customer data in-country» settingMicrosoft Learn: Flex routing (EU and EFTA)AWS: Model support by AWS Region in Amazon BedrockGoogle Cloud: Data residency for generative AIFINMA: Supervisory notice 08/2024 on governance and risk management when using AIFINMA: Circular 2018/3 OutsourcingFedlex: Financial Institutions Ordinance, Art. 17 (outsourcing)Fedlex: Financial Institutions Act, Art. 69 (professional secrecy)Fedlex: Banking Act, Art. 47 (banking secrecy)Phoeniqs: Active ModelsPhoeniqs: Preview Models (Kimi K3, GLM 5.3, DeepSeek V4.1 Flash)Phoeniqs: Release note on the model portfolio, May 2026Artificial Analysis: Intelligence Index (methodology and values, v4.3.2)Artificial Analysis: GPT-4o (value estimated)Artificial Analysis: GPT-5.1 (value estimated)Artificial Analysis: gpt-oss-120bIndex values of all models (modelgrep, data from Artificial Analysis)Infomaniak: open-source modelsSwisscom Swiss AI Platform: model overviewAlpineAI: SwissGPTMicrosoft Learn: region availability of Azure OpenAI models (Data Zone EU)OpenAI: GPT-6 and EU data residency in the APIOpenAI: data and inference residency for ChatGPT EnterpriseGoogle Cloud: data residency of models on Vertex AIMeta Model API: Geographic Use Policy (Muse Wonder)
Choose up to three options. Data path, performance and cost sit next to each other, row by row.
Artificial Analysis Intelligence Index. Independent benchmark by Artificial Analysis. All models solve the same ten demanding task sets: agentic tasks and real office work (30%), general knowledge, long documents and hallucinations (30%), coding (20%) and scientific reasoning (20%). Scale 0 to 100, higher is better. The tests are deliberately very hard: the best model scores 58 today.
Version 4.3.2, values as of 23 October 2026. "ca." = estimated by Artificial Analysis, the independent measurement is still pending.
Your selection is already filled in:
Cloud AI is usually commissioned processing under the DPA, and responsibility stays with you.
Even when outsourcing to the cloud, you remain fully responsible to your supervisor and your clients.
Sources: FDPIC: Cross-border transfer of personal dataFDPIC: Outsourcing of data processingFINMA: Supervisory notice 08/2024 on governance and risk management when using AIFINMA: Circular 2018/3 OutsourcingFedlex: Financial Institutions Ordinance, Art. 17 (outsourcing)Fedlex: Financial Institutions Act, Art. 69 (professional secrecy)
Humanyx accompanies you as a partner from choosing the operating route to ongoing operations of your Swiss pipeline. Contract and governance are included. You get finished AI employees for your processes instead of an infrastructure project.