AI on Demandpowered by Swiss AI Initiative

Apertus — the Swiss AI Initiative’s fully open language model: developed by ETH Zurich, EPFL and CSCS, with 8 billion parameters and support for over 1,000 languages. The model weights, training data and methods are fully documented — operated by stepping stone on Swiss infrastructure.

Apertus is a language model developed as part of the Swiss AI Initiative by ETH Zurich, EPFL and the Swiss National Supercomputing Centre (CSCS). It is fully open-source — not only are the model weights available, but the training data, code and methods are also documented and reproducible.

With 8 billion parameters and support for over 1,000 languages, Apertus is a versatile, general-purpose model for text, analysis and agent-based workflows. It has been specifically trained to meet the requirements of the EU AI Act: opt-out requests are respected, personal data is removed and memorisation is avoided. stepping stone operates Apertus on Swiss infrastructure — Swiss AI, operated in Switzerland.

Companies, public authorities and organisations that require a high-performance language model and prioritise maximum transparency, compliance and digital sovereignty. Particularly suitable for regulated industries and the public sector.

Typical areas of application: text generation, summarisation and analysis in over 1,000 languages; agent-based workflows with tool integration; assistance systems for multilingual organisations; and AI applications with heightened compliance requirements (EU AI Act, data protection).

Fully open-source (Apache 2.0). Swiss research. Swiss data centres. Compliant with the EU AI Act.

Apertus is the only fully open-source language model developed by Swiss researchers — created by ETH Zurich, EPFL and CSCS. Over 1,000 languages, 65,536-token context windows, tool support for agent-based workflows. Trained on 15 trillion tokens using fully documented data and methods. Personalised consultancy and operation provided by stepping stone in Bern.

Your Consultant

Scope of services

Swiss AI on demand

Access to Apertus for text generation, analysis and agent-based workflows. Over 1,000 languages, a 65,536-token context window, fully open-source and compliant with the EU AI Act.

GPU performance on demand

Scalable computing power on Swiss infrastructure. Efficient and transparent — you pay as you go.

Managed service

Deployment, monitoring, maintenance and support on Swiss infrastructure, with personalised advice. stepping stone takes care of the day-to-day running so that you can focus on the benefits.

Areas of application

Compliance & Regulation

Apertus is the only fully open language model designed from the ground up to comply with the EU AI Act.

Regulatory authorities and regulated organisations use it in situations where transparency, traceability and data protection are non-negotiable. Opt-out requests are respected, personal data is removed, and data retention is actively avoided.

Multilingual applications

With over 1,000 languages and 65,536 token context windows, Apertus covers virtually every multilingual application.

Organisations with international teams or multilingual document collections use it for text generation, summarisation and analysis. The tool supports agent-based workflows — fully open, reproducible and run on Swiss infrastructure.

Benchmark

The benchmarks were measured using the vllm bench tool against the production API gateway. The standard input sizes were 1,024 tokens for input and 256 tokens for output, which corresponds to 2–3 book pages or 500–750 words.

If necessary, higher input sizes can be set.

 

Call

# Set your personal key:
STONEY_KEY=sk-...

# Make key visible for vllm bench:
export OPENAI_API_KEY=$STONEY_KEY

# Start the benchmark
vllm bench serve \
  --backend openai-chat  \
  --model "apertus-ai/apertus-ai/Apertus-v1.5-8B" \
  --base-url llm.stoney-cloud.com \
  --endpoint /v1/chat/completions \
  --dataset-name random \
  --random-input-len 1024 \
  --random-output-len 256
  --num-prompts 50 \
  --max-concurrency 1
  --tokenizer "Qwen/Qwen2.5-7B-Instruct"
  --percentile-metrics e2el

 

Result

============ Serving Benchmark Result ============
Successful requests:                     48        
Failed requests:                         2         
Maximum request concurrency:             1         
Benchmark duration (s):                  156.89    
Total input tokens:                      79041     
Total generated tokens:                  12288     
Request throughput (req/s):              0.31      
Output token throughput (tok/s):         78.32     
Peak output token throughput (tok/s):    257.00    
Peak concurrent requests:                2.00      
Total token throughput (tok/s):          582.11    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          3138.17   
Median E2EL (ms):                        3137.92   
P99 E2EL (ms):                           3175.76   
==================================================

 

Legend

  • Successful requests: Successful prompt requests.
  • Failed requests: Unsuccessful prompts.
  • Maximum request concurrency: How many requests the model processes simultaneously.
  • Benchmark duration (s): The duration of the benchmark run in seconds.
  • Total input tokens: The total number of input tokens.
  • Total generated tokens: The total number of tokens generated by the model.
  • Request throughput (req/s): The number of requests processed per second.
  • Output token throughput (tok/s): The average number of tokens generated per second.
  • Peak output token throughput (tok/s): The maximum measured number of output tokens per second.
  • Peak concurrent requests: The maximum measured number of requests processed simultaneously.
  • Total token throughput (tok/s): The average of all tokens processed during the measurement.
  • Mean End-to-End Latency (E2EL) (ms): The average time, in milliseconds, taken by a complete request-response cycle from input to response.
  • Median E2EL (ms): The expected time taken to fully generate and return a response.
  • p99 E2EL (ms): The p99 E2EL is the slowest 99 % of response times.
  • Tokenizer: The tokenizer is used to send queries to the evaluated model during a benchmark. These are typically small, publicly available models, such as Qwen/Qwen2.5-7B-Instruct.

Price

ModelContext lengthInput/MTokOutput/MTok
Apertus-v1.5-8B256k0.12000.5800
All prices are in CHF/MTok, excluding VAT.