Benchmark: search marketplace on willhaben with an AI agent
Wield's recipe passed 9 of 9 runs on willhaben, with a median of 0.6 s and no model cost. A Claude browser agent passed 4 of 4 runs, with a median of 162.0 s and $0.47 per run. The benchmark did not flag any agent run as blocked. These results are preliminary, as of 7 November 2026.
willhaben Search the willhaben marketplace (Austria) for second-hand items
Read-only, no loginPasses the daily check
Parameters (6)
| Metric | Wield recipe | Browser Use + Claude Sonnet 5.5 |
|---|---|---|
| Success | 100% | 100% |
| Median time | 582 ms | 162 s |
| Cost per run | $0 | $0.472 |
9 runs each, 7 November 2026. How we measure
The task
Each side ran the same three searches. A run passes when it returns at least five listings with a title and a price.
| Query | Parameters |
|---|---|
| q1 | query = fahrrad, min_price = 50, max_price = 300, state = Wien, sort = 3 |
| q2 | query = kinderwagen, state = Oberösterreich |
| q3 | query = ps5, max_price = 400 |
Results
| Wield recipe | Browser agent | |
|---|---|---|
| Runs passing the check | 9 of 9 | 4 of 4 |
| Median time per run | 582 ms | 162.0 s (162,030 ms) |
| Model cost per run | $0 | $0.47 |
| Requests per run (median) | 1 | 204 (to the site: 22) |
| Blocked by the site | 0 runs | 0 of 4 runs |
The browser agent took a median of 12 steps per run, with 83,835 input tokens and 3,826 output tokens. Its median run made 204 requests, and 22 of them reached willhaben.at.
Why the results differ
The browser agent is the costly side here. It read each page in turn, and the token count per run is high. Wield makes one search request from the user's own browser and runs no model.
Wield's slowest raw run hit the 60 s run limit, which the log records as a timeout. The summary counts include all 9 runs as passing. Read the timeout as one raw run that did not finish in time.
Caveats
- Very small agent sample. The browser agent has 4 runs, and Wield has 9. The benchmark is still running and is paced at about 30 site requests per hour.
- No blocking here. This task shows speed and cost. It does not show access, because no agent run was blocked.
- Human pace only. Wield runs at human pace in the user's own browser. If a check appears, Wield stops with
site_challengeand the user solves it.
Terms of willhaben
willhaben's terms prohibit automated reading of ads. The recipe makes one search page at a time that the user starts, from the user's own browser. Contacting sellers stays on willhaben.at.
FAQ
Does willhaben block browser agents?
Not in our runs. None of the 4 browser agent runs was marked blocked. The agent cost $0.47 per run, the highest of the marketplace tasks in this set.
How fast is the Wield recipe for search marketplace on willhaben?
In our benchmark as of 7 November 2026, the median run took 0.6 s. The fastest run took 0.3 s and the slowest 60.1 s.
Can I rerun this benchmark?
The method and per-task rows are on the benchmark page. The raw logs hold destinations, timing, tokens and check results, not page contents.