Performance and stress testing in k6, also for LLM applications
Find out how many users your application can take, how model response time grows under load and whether token limits really work, before a marketing campaign checks it for you.
Book a free consultationWho it is for
- Applications with a chatbot or agent (LLM cost and latency)
- E-commerce before peak season
- Startups before launch
- Applications with an SLA (DORA)
Types of tests
- Load, stress, spike and soak
- For LLM: time to first token (TTFT) and full response time, p95
- For LLM: behavior of queues and the model provider’s rate limits
- For LLM: token cost per 1000 conversations
- For LLM: resilience to very long prompts
How the testing works
k6 with scripts in TypeScript or JS, scenarios based on real paths, thresholds (p95, errors, cost), application-side monitoring where available.
You designate the environment, and production is tested only with consent and in an agreed window. LLM endpoints are tested with a model mock or within an agreed budget.
What you get
- k6 scripts that are your property
- A report with charts
- Bottlenecks and recommendations
- Thresholds to wire into CI
Packages
| Package | What it includes | Time |
|---|---|---|
| Performance Testing | k6 scenarios (including LLM endpoints), execution, report with thresholds | approx. 5 working days |
| Performance in CI | k6 thresholds wired into the pipeline | on request |
Frequently asked questions
Will load tests burn our model API budget?
LLM endpoints are tested with a model mock or within a budget agreed up front. There is no load on a production key without consent.
Is production tested?
Only with consent and in an agreed window. By default testing runs on the environment you designate.
Do the scripts stay with us?
Yes. The k6 scripts are your property, and the thresholds can be wired into CI.
Contact
A few sentences are enough: what your application does, where it uses AI and what you want tested. The reply comes within 1 working day, with a proposed time for a free call.