Benchmark: search practitioners on Doctolib with an AI agent
On Doctolib, Wield's recipe passed 9 of 9 practitioner searches with a median of 891 ms and no model cost per run. A browser agent running Claude Sonnet 5.5 passed 2 of 2 runs, with a median of 160 seconds and $0.38 per run. The agent side has only two runs, so treat it as an early signal. Neither side was blocked by the site.
Doctolib Search doctors on Doctolib by specialty and city
Read-only, no loginRuns in your browser only
Parameters (4)
| Metric | Wield recipe | Browser Use + Claude Sonnet 5.5 |
|---|---|---|
| Success | 100% | 100% |
| Median time | 891 ms | 160 s |
| Cost per run | $0 | $0.376 |
9 runs each, 7 November 2026. How we measure
The task
Both sides ran the same three searches, for specialty and city.
| Query | Parameters |
|---|---|
| q1 | specialty = Hautarzt, city = Berlin |
| q2 | specialty = Zahnarzt, city = München |
| q3 | specialty = Kinderarzt, city = Hamburg |
Pass check: at least five doctors, each with a name and a city.
Results (7 Nov 2026)
| Wield recipe | Browser agent | |
|---|---|---|
| Runs passing the check | 9 of 9 | 2 of 2 |
| Median time per run | 891 ms | 160.5 s |
| Model cost per run | $0 | $0.38 |
| Requests per run (median) | 3 | 688.5 (34 to the site) |
| Blocked by the site | 0 runs | 0 of 2 runs |
The browser agent took a median of 19.5 steps, 118,894 input tokens and 5,536 output tokens per run. Its log records no HTTP 401, 403 or 429 response from the site. Wield's fastest run took 216 ms and its slowest 24,801 ms.
The raw logs also hold one extra Wield entry that stopped with a tab_error. The extension could not load its content script in the site tab. The table uses the summary counts, which show 9 of 9 passing.
Why the results differ
The recipe reads the practitioner data that the search page embeds. It returns names, specialties, addresses, whether new patients are accepted, video consultation, and the ids that the slot lookup needs. It makes three requests per run at the median, from a doctolib.de tab in your browser, one search at a time.
The browser agent opens the search page, changes filters and reads results. Each of its median 19.5 steps waits for a model call.
Caveats
- Small and preliminary. The benchmark was still running on 7 Nov 2026. The agent side has 2 runs. Pacing at about 30 site requests per hour keeps counts low on some sides.
- Sensitive results. Health searches stay on your device. Wield does not send them to wield.run.
- Cost is model cost only. Wield's $0 covers model cost. No model runs in the loop when a recipe runs.
Terms of Doctolib
Doctolib's terms prohibit automated access and scraping. The recipe runs one search that you start, from your own browser, with three requests. Do not loop it to collect practices in bulk. Check the site's terms yourself. Wield's recipes note known restrictions.
FAQ
Does Doctolib block browser agents?
Not in these runs. The browser agent was blocked by the site in 0 of 2 runs, and its log records no 401, 403 or 429 response. Other specialties, cities and times may differ, and bot protection can change at any time.
How fast is the Wield recipe for search practitioners on Doctolib?
The median run took 891 ms over 9 runs. The fastest took 216 ms and the slowest 24,801 ms. All 9 runs passed the check in the summary.
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.