Home

AI on Demandpowered by DeepSeek

DeepSeek-V4-Flash-0731 — a high-capacity open-weight Mixture-of-Experts model for coding, reasoning and agentic workflows, operated by stepping stone entirely on Swiss infrastructure.

DeepSeek-V4-Flash-0731 is an open-weight language model designed for demanding coding, reasoning and agentic workloads. Its Mixture-of-Experts architecture combines a large overall model capacity with efficient inference by activating only part of the model for each token.

The 0731 release places particular emphasis on agentic capabilities. It can work with tools, process multi-step tasks and support workflows in which the model has to analyse intermediate results and decide what to do next. DeepSeek reports strong performance on coding-agent, terminal and tool-use benchmarks.

stepping stone operates DeepSeek-V4-Flash-0731 entirely on Swiss infrastructure. Customers access the model through an OpenAI-compatible API and can integrate it directly into existing applications, development environments and agentic systems.

DeepSeek-V4-Flash-0731 is suitable for organisations that need a capable model for complex technical workloads without transferring their data to proprietary US-based AI platforms.

Typical use cases include software development, debugging, agentic workflows and technical assistans to name a few.

Open Weights. Swiss infrastructure. Powerful agentic capabilities.

DeepSeek-V4-Flash-0731 combines high model capacity with an architecture designed for efficient inference. Its strengths in coding, tool use and multi-step workflows make it particularly suitable as a foundation for agentic applications.

Operation by stepping stone keeps data processing on Swiss infrastructure and provides customers with a standardised API, predictable operation and personal support from our team in Bern.

Your Consultant

Scope of services

AI model on demand

Access to DeepSeek-V4-Flash-0731 for coding, analysis, reasoning and agentic workflows through an OpenAI-compatible API.

GPU performance on demand

Scalable GPU infrastructure in Swiss data centres for demanding inference workloads, without the need to build and operate dedicated AI infrastructure internally.

Managed service

Deployment, monitoring, maintenance and support by stepping stone. We operate the model and infrastructure so customers can focus on integrating AI into their applications and workflows.

Areas of application

Software development

DeepSeek-V4-Flash-0731 is particularly well suited to demanding software-development workflows.

The model can analyse existing code, work across repositories, assist with debugging and support multi-step development tasks. Combined with an agent harness and suitable tools, it can iteratively inspect files, modify code, run tests and react to the results.

Agentic workflows

DeepSeek-V4-Flash-0731 is designed for workflows in which the model does more than generate a single response.

Through tool calling, the model can interact with external functions and systems, evaluate intermediate results and continue working through a task over several steps. This makes it suitable for technical agents, automated analysis workflows and other applications that require autonomous execution. DeepSeek's current API documentation lists tool-call support for the Flash model.

Analysis and reasoning

For tasks that require more than simple extraction or text generation, DeepSeek-V4-Flash-0731 provides a high-capacity option for technical analysis, structured problem solving and complex multi-step reasoning.

This makes it suitable where smaller general-purpose models no longer provide sufficient reliability, while still allowing the model to be operated within a controlled Swiss infrastructure.

Benchmark

The benchmarks were measured with a python script.
For a fixed resolution the time per image is measured, which is needed for generating the image.
Because the required time depends on resolution and number of steps different configurations can be tested.
Step-by-step instructions and the required Python script can be downloaded from GitHub.


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 "deepseek-ai/DeepSeek-V4-Flash-0731" \
 --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-mode deepseek_v4 \
 --percentile-metrics e2el


Result

============ Serving Benchmark Result ============
Successful requests:                     50        
Failed requests:                         0         
Maximum request concurrency:             1         
Benchmark duration (s):                  118.57    
Total input tokens:                      55350     
Total generated tokens:                  12800     
Request throughput (req/s):              0.42      
Output token throughput (tok/s):         107.96    
Peak output token throughput (tok/s):    173.00    
Peak concurrent requests:                2.00      
Total token throughput (tok/s):          574.79    
----------------End-to-end Latency----------------
Mean E2EL (ms):                          2371.00   
Median E2EL (ms):                        2288.91   
P99 E2EL (ms):                           3317.25   
==================================================


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 between input and completed output.
  • Median E2EL (ms): The typical time between input and completed output.
  • p99 E2EL (ms): The "worst case" time until completed output.

Price

ModelInput/MTokOutput/MTok
DeepSeek-V4-Flash-07310.4500   1.8000
All prices are in CHF/MTok, excluding VAT.

Wir verwenden Cookies, um unsere Website und unseren Service zu optimieren. Weitere Informationen findest du in unserer Datenschutzerklärung.