Benchmark: search trips on FlixBus with an AI agent
On FlixBus, Wield's recipe found trips in 9 of 9 runs, with a median of 0.8 seconds and no model cost. A browser agent (Browser Use driving Claude Sonnet 5.5) passed 3 of 3 runs, with a median of 41 seconds and $0.23 per run. Both sides ran the same three route queries and the same pass check on 7 November 2026.
FlixBus Search bus trips between two cities on a date
Read-only, no loginPasses the daily check
Parameters (5)
| Metric | Wield recipe | Browser Use + Claude Sonnet 5.5 |
|---|---|---|
| Success | 100% | 100% |
| Median time | 791 ms | 41 s |
| Cost per run | $0 | $0.227 |
9 runs each, 7 November 2026. How we measure
The task
Each side ran the same three searches. Each search asks for trips on one route and one date.
| Query | Parameters |
|---|---|
| q1 | from = Berlin, to = München, date = 2026-11-20 |
| q2 | from = Hamburg, to = Köln, date = 2026-11-18 |
| q3 | from = Frankfurt am Main, to = Stuttgart, date = 2026-11-25 |
Pass check: at least one trip departs on the requested date, and the result shows departure, arrival and price.
Results (7 November 2026)
| Wield recipe | Browser agent | |
|---|---|---|
| Runs passing the check | 9 of 9 | 3 of 3 |
| Median time per run | 791 ms | 41.1 s (41,150 ms) |
| Model cost per run | $0 | $0.23 |
| Requests per run (median) | 3 | 251.5 (to the site: 55.5) |
| Blocked by the site | 0 runs | 0 of 3 runs |
Wield's fastest run took 539 ms and its slowest took 8.2 s. The recipe had no failed runs.
Why the results differ
The recipe sends the same requests that the FlixBus search page makes. It runs from a FlixBus tab in your own browser, at human pace, with three requests per run. No model sits in the loop, so a run uses no tokens.
The browser agent works differently. It loads pages, reads them and clicks through the search form. A median run took 9.5 steps and used 67,944 input tokens and 1,472 output tokens. It sent about 250 requests per run, and most of them went to hosts other than FlixBus.
The site did not block the browser agent in these runs. The log records no HTTP 401, 403 or 429 errors. The gap is in time and cost, not in access.
Caveats
- Small sample. The browser agent passed 3 of 3 runs. Treat its numbers as a rough guide.
- Preliminary. The benchmark was still running on 7 November 2026, so the numbers will change.
- Paced. Runs are paced at about 30 site requests per hour, so counts are low on some sides.
- Model cost only. Wield's $0 covers model cost. It does not count your time or the browser you already use.
Terms of FlixBus
FlixBus's terms restrict automated extraction of timetable and price data. The recipe runs one search that you start, from your own browser, with three requests. Booking stays in the FlixBus shop, and the recipe does not book for you. Check the site's terms before you automate any use of it.
FAQ
Does FlixBus block browser agents?
Not in these runs. The browser agent recorded no HTTP 401, 403 or 429 errors from FlixBus. Three runs on three routes, on one day, cannot show how the site treats agents in general.
How fast is the Wield recipe for search trips on FlixBus?
The median run took 791 ms across nine runs. The fastest took 539 ms and the slowest took 8.2 s. The browser agent median on the same three queries was 41.1 s.
Can I rerun this benchmark?
The method, the task file and the raw logs are described on the benchmark page. The logs record destinations, timing, tokens and checks. They hold no page contents.