Benchmark: search stays on Airbnb with an AI agent
On Airbnb, Wield's recipe passed 8 of 9 search runs with a median of 1.4 seconds and no model cost per run. A browser agent running Claude Sonnet 5.5 passed 4 of 4 runs, with a median of 70 seconds and $0.19 per run. Here the agent had the higher pass rate. Wield's one miss has no error text.
Airbnb Search Airbnb stays for a place and dates
Read-only, no loginRuns in your browser only
Parameters (6)
| Metric | Wield recipe | Browser Use + Claude Sonnet 5.5 |
|---|---|---|
| Success | 89% | 100% |
| Median time | 1.4 s | 70 s |
| Cost per run | $0 | $0.194 |
9 runs each, 7 November 2026. How we measure
The task
Both sides ran the same three searches, priced in EUR.
| Query | Location | Check-in | Check-out | Adults |
|---|---|---|---|---|
| q1 | Lisbon | 4 Dec 2026 | 7 Dec 2026 | 2 |
| q2 | Paris, France | 11 Dec 2026 | 13 Dec 2026 | 2 |
| q3 | Barcelona | 15 Jan 2027 | 18 Jan 2027 | 1 |
Pass check: at least five stays, each with a name and a price.
Results (7 Nov 2026)
| Wield recipe | Browser agent | |
|---|---|---|
| Runs passing the check | 8 of 9 | 4 of 4 |
| Median time per run | 1.4 s | 70.2 s |
| Model cost per run | $0 | $0.19 |
| Requests per run (median) | 1 | 361 (19 to the site) |
| Blocked by the site | 0 runs | 0 of 4 runs |
The browser agent took a median of 10 steps, 80,299 input tokens and 2,688 output tokens per run. Its log records no HTTP 401, 403 or 429 response from the site. Wield's fastest run took 1,002 ms and its slowest took 5,445 ms.
Wield missed one run. The log holds no error text for it, and the table counts it as a miss.
Why the results differ
The Wield recipe makes one request per search at the median. It sends that request from an airbnb.com tab in your own browser, at human pace. The browser agent loads pages, reads them and clicks through. Each of its median 10 steps waits for a model call, so a run takes far longer than one request.
Caveats
- Small and preliminary. The benchmark was still running on 7 Nov 2026. The browser agent side has 4 runs. Pacing at about 30 site requests per hour keeps counts low on some sides.
- Wield's miss is part of the data. Wield passed 8 of 9 here, and the table shows it.
- Live numbers. The live benchmark row on this page comes from
bench/results.json.
Airbnb's terms, in plain words
Airbnb's robots.txt disallows AI training crawlers (ClaudeBot, anthropic-ai). Airbnb's terms prohibit scraping. Wield runs the recipe only when you ask, one search at a time from your own browser, with one request per search. The scope is set: you start each search, one search per call, from your own browser, with no bulk collection and no purchases. Booking stays on airbnb.com. Check the site's terms yourself. Wield's recipes note known restrictions.
FAQ
Does Airbnb block browser agents?
Not in these runs. The browser agent was blocked by the site in 0 of 4 runs, and its log records no 401, 403 or 429 response. Other tasks, dates and times may give different results, and bot protection can change at any time.
How fast is the Wield recipe for search stays on Airbnb?
The median run took 1.4 seconds (1,374 ms) over 9 runs. The fastest run took 1,002 ms and the slowest 5,445 ms. The recipe passed 8 of those 9 runs.
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.