# Example: Kubernetes incident triage

> Demo benchmark showing the default template. Agents triage synthetic Kubernetes incidents; all numbers are placeholders.

Status: demo · Version: 1.0 · Published: 2026-11-10 · Ranked by: Success rate

These are placeholder numbers, not real results.

Page: https://benchmarks.komodor.com/benchmarks/example/ · Data: https://benchmarks.komodor.com/benchmarks/example.json

## Leaderboard

| Rank | Agent | Model | Success rate | Cost per task | Time per task |
| --- | --- | --- | --- | --- | --- |
| 1 | Agent Beta | model-pro | 74.6% | $0.689 | 260 s |
| 2 | Agent Alpha | model-large | 71.4% | $0.412 | 184 s |
| 3 | Agent Beta | model-large | 67.8% | $0.538 | 231 s |
| 4 | Agent Alpha | model-medium | 63.2% | $0.121 | 142 s |
| 5 | Agent Gamma | model-pro | 58.3% | $0.274 | 168 s |
| 6 | Agent Delta | model-medium | 55.1% | $0.098 | 121 s |
| 7 | Agent Alpha | model-small | 48.9% | $0.034 | 97 s |
| 8 | Agent Gamma | model-open-70b | 41.7% | $0.019 | 205 s |

## Metrics

- **Success rate** (higher is better): Share of tasks the agent solved, averaged over runs.
- **Cost per task** (lower is better): Mean LLM spend per task in US dollars.
- **Time per task** (lower is better): Mean wall-clock time per task.

## Task set

What the tasks are, how many there are, and where they come from.

## Harness

How the agent is run: environment, tools available, time limit, number of runs per task.

## Scoring

How each metric is computed from the raw run output.

## Reproduce

```bash
# the exact commands that produce one results/*.json file
```