Benchmark: search cars on AutoScout24 with an AI agent
On AutoScout24, Wield's recipe passed 9 of 9 car searches with a median of 0.9 seconds and no model cost per run. A browser agent running Claude Sonnet 5.5 passed 5 of 5 runs, with a median of 144 seconds and $0.54 per run. Neither side was blocked by the site in these runs.
AutoScout24 Search used cars by make, model, price, year and mileage
Read-only, no loginPasses the daily check
Parameters (11)
| Metric | Wield recipe | Browser Use + Claude Sonnet 5.5 |
|---|---|---|
| Success | 100% | 100% |
| Median time | 860 ms | 144 s |
| Cost per run | $0 | $0.539 |
9 runs each, 7 November 2026. How we measure
The task
Both sides ran the same three searches with price and age filters.
| Query | Filters |
|---|---|
| q1 | make = volkswagen, model = golf, price up to 15,000, first registered from 2018, sort by price |
| q2 | make = bmw, model = 3er, price up to 20,000, max 100,000 km |
| q3 | make = tesla, model = model-3, price up to 30,000 |
Pass check: at least five cars, each with a title and a price.
Results (7 Nov 2026)
| Wield recipe | Browser agent | |
|---|---|---|
| Runs passing the check | 9 of 9 | 5 of 5 |
| Median time per run | 860 ms | 144.5 s |
| Model cost per run | $0 | $0.54 |
| Requests per run (median) | 1 | 405 (38 to the site) |
| Blocked by the site | 0 runs | 0 of 5 runs |
The browser agent took a median of 23 steps, 194,262 input tokens and 6,258 output tokens per run. Its log records no HTTP 401, 403 or 429 response from the site. Wield had no failed runs. Its fastest run took 391 ms and its slowest 9,344 ms.
Why the results differ
The Wield recipe reads the search data that the results page already embeds. It honours the sort, price and first-registration filters, and it makes one request per search at the median. It runs from an autoscout24.de tab in your own browser, at human pace.
The browser agent needed a median of 23 steps to set filters, read results and move between pages. Each step waits for a model call. That is why its runs took over two minutes at the median.
Caveats
- Small and preliminary. The benchmark was still running on 7 Nov 2026. Pacing at about 30 site requests per hour keeps counts low on some sides.
- Cost is model cost only. Wield's $0 covers model cost. No model runs in the loop when a recipe runs.
- Live numbers. The live benchmark row on this page comes from
bench/results.json.
AutoScout24's terms, in plain words
AutoScout24.de's robots.txt disallows AI training crawlers (ClaudeBot). AutoScout24's terms prohibit automated data collection. Wield runs the recipe only when you ask, one search page at a time from your own browser. The scope is set: you start each search, one search per call, from your own browser, with no bulk collection and no purchases. Contacting dealers stays on autoscout24.de. Check the site's terms yourself. Wield's recipes note known restrictions.
FAQ
Does AutoScout24 block browser agents?
Not in these runs. The browser agent was blocked by the site in 0 of 5 runs, and its log records no 401, 403 or 429 response. Other filters and times may differ, and bot protection can change at any time.
How fast is the Wield recipe for search cars on AutoScout24?
The median run took 860 ms over 9 runs. The fastest took 391 ms and the slowest 9,344 ms. All 9 runs passed the check.
Can I rerun this benchmark?
Yes. The method is on the benchmark page, and the query table above is the full input. The raw per-run logs keep destinations, timing, tokens and checks. They do not keep page contents.