Live on every query.
Not a screenshot.
Every map search below calls Baidu's overseas place API in real time. The AI Search tab replays a captured 10-query benchmark, field by field, so you can audit the method.
Live API · place_abroad/v1/search · no cache, no sample
Search the map, audit the benchmark.
Live · place_abroad/v1 · every query
Type a place, a category or a street, then pick a city. The request goes to Baidu Maps now; the fields below are the response, including duplicates and empty values — flagged, not hidden.
Results
—A fresh call goes to place_abroad/v1 on every search. Nothing is read from a stored sample.
The call succeeded but nothing matched. Try a broader term or another city.
Replay · 10 queries · 54 calls · 23–25 Oct 2026
Select a use case to see what Baidu's evidence layer and Google's AI Mode returned in the controlled run — then move the volume slider to price the architecture. These ten cases are a stored capture, not a live call.
Price the retrieval layer
Move the slider to compare Baidu's overseas turbo list price and a 50% scenario with Gemini grounding. Token charges are excluded on both sides.
The 50% rate is a scenario, not a quote. Baidu includes 1,500 free calls per month; Gemini grounding is priced at $14 per 1,000 searches plus tokens.
Method: ten fixed English queries, 20 calls in the controlled round, 30 in the recheck and 4 immediate reruns. Google organic and AI Mode were observed through third-party SERP APIs; Baidu was called directly. Ten probes are a signal, not a market-share estimate.
A live call, a server-side key, nothing hidden.
Key stays on the server
The browser calls our own /api/places route. The Baidu Maps key lives in the BAIDU_MAP_AK environment variable — it is never written into this page, a request URL or a log.
No cache, no sample
Each search is a fresh upstream call, and the latency and timestamp are printed under the results. If the key is missing, the page says so instead of showing old data.
Flaws stay visible
Duplicates, missing ratings, unlabelled prices and missing hours are kept and flagged. That is what a field evaluation is for.
Caveat: place coverage, ranking and field completeness vary by city and query, and the API may change without notice. The AI Search benchmark is a stored capture of ten probes, not a population error rate. Test your own market before quoting a customer.
Run the same test on your own queries.
Register an international Baidu account, then send us the use cases that matter. We will run them, share the raw fields and scope a pilot with free test credits.