The Pulse
Apertus 1.5 Adds Swiss-Hosted APIs From Four Providers
The Swiss AI Initiative’s Apertus 1.5 release notes detail developer API access from Swisscom, Infomaniak, PHOENIQS and Safe Swiss Cloud. The 8B and 70B models also remain available as open weights, with the 70B version offered through all
AI.info Team · Written with AI ·
“Switzerland shouldn’t only consume AI, it should be able to build it. Apertus 1.5 shows the Swiss research community can deliver a genuinely open, genuinely capable model — and Swisscom’s job is to make it usable: hosted in Switzerland, hardened for regulated industries, available from day one.”
Sarah Levy, Head of Swiss AI Platform
The Swiss AI Initiative’s October 15 release notes put a developer-facing network around Apertus 1.5: access to the model through APIs from Swisscom, Infomaniak, PHOENIQS and other providers. Four services in the initiative’s September general-availability guide offer metered access to the 70B model with data processing in Swiss data centers. The update gives developers alternatives to downloading and operating the model themselves.
Four metered APIs, one Swiss data-location claim
The September guide lists Swisscom, Infomaniak, PHOENIQS and Safe Swiss Cloud as offering the 70B model on a per-token basis from Swiss data centers. That location describes where user data is processed, according to the guide; it is not a blanket statement about every service connected to Apertus. The guide also lists OnPrem.ai for 70B deployments and stepping stone for 8B, while Public AI and Featherless offer access from various regions.
The distinction matters for organizations choosing where to send prompts. Some listed services are hosted API offerings, while others provide on-premise deployment or model access through services with different locations and terms. The initiative’s guide identifies provider options, but does not publish a single shared price, service-level agreement or data-retention policy for the ecosystem; those details require checking with each provider.
The weights remain open for self-hosting
Apertus 1.5 is available in 8B and 70B parameter versions. The weights reached Hugging Face on July 24, 2026, under the Apache 2.0 license, months before the October release notes described the expanding service network. Developers can use provider APIs or download the weights and run them on their own systems.
The release extends Apertus 1.0 through continued pretraining: the 8B model received 4 trillion additional tokens, and the 70B model received 2 trillion, bringing their training totals to 19 trillion and 17 trillion tokens respectively. The initiative says it released training data and details alongside the weights. That gives teams choosing self-hosting access to more than model files, though running the 70B version calls for server-class hardware.
Images, experimental audio and a 262,144-token window
Apertus 1.5 adds image and audio understanding, an optional thinking mode, improved instruction-following and tool use. Its context window reaches 262,144 tokens, four times the length of Apertus 1.0. Audio input remains experimental, and the models generate text rather than images or audio.
The new capabilities also have deployment implications: the initiative notes that some require configuration changes and inference-side upgrades. Providers need to support the relevant model features for customers to use them through an API. A listed endpoint alone does not establish that every feature is enabled or behaves identically across services.
Availability is broader than the benchmark evidence
The initiative’s September guide reports a preliminary comparison of API performance using simulated workloads, with providers anonymized and results varying across latency and token throughput. It says the results are early and should be reproduced independently before users rely on them to choose a service. The figures do not identify which anonymized provider corresponds to any named Swiss company.
For now, Apertus 1.5’s service ecosystem offers a practical choice: metered Swiss-hosted APIs from several providers, deployments on private infrastructure, and access through services operating in other regions. Its open weights keep those options from becoming the only route to the model. The October 15 release notes make the concrete change clear: developers can use Apertus 1.5 through a growing provider list, while the model itself remains available to download.