Benchmarks
X (Twitter) API benchmarks: live latency, success rate and schema coverage.
Measured on real customer calls over the last 14 days, not on a lab run. The raw numbers are public, and the script below lets you check them with your own key.
- p50 latency
- 507 ms
- p95 latency
- 2,399 ms
- HTTP 200 rate
- 100%
- Calls measured
- 107
14-day window, computed 2026-11-19. End-to-end time of HTTP 200 responses on paid data and account-action calls. Raw telemetry JSON
Scenario A
Keyword and cashtag search
Search with the X advanced search syntax (keywords, $cashtags, from:, since:, min_faves:). Each call returns up to 20 tweets with 30 fields each, including the author's profile (34 fields), so no second call is needed to know who posted.
Per call: Up to 20 tweets, $0.0006.
| Endpoint | Calls | p50 | p95 | HTTP 200 |
|---|---|---|---|---|
| GET /twitter/tweet/advanced_search | 0 | Not enough customer calls yet | ||
Scenario B
Profile and timeline ingestion
One profile per lookup (25 fields), up to 20 profiles per batch call by ID, and a user's latest tweets 20 at a time.
Per call: 1 profile, 20 profiles, or up to 20 tweets, $0.0006.
| Endpoint | Calls | p50 | p95 | HTTP 200 |
|---|---|---|---|---|
| GET /twitter/user/info | 23 | 130 ms | 463 ms | 100% |
| GET /twitter/user/batch_info_by_ids | 21 | 398 ms | 518 ms | 100% |
| GET /twitter/user/last_tweets | 42 | 1,709 ms | 3,045 ms | 100% |
Scenario C
Threads and replies for RAG
The replies under a tweet, sorted by relevance, time or likes, and the whole conversation a tweet belongs to, as context for a language model.
Per call: Up to 20 tweets, $0.0006.
| Endpoint | Calls | p50 | p95 | HTTP 200 |
|---|---|---|---|---|
| GET /twitter/tweet/replies/v2 | 0 | Not enough customer calls yet | ||
| GET /twitter/tweet/thread_context | 0 | Not enough customer calls yet | ||
How it compares.
We publish latency only for our own API, because we can only measure our own traffic. Cost is at list price for 100,000 tweets from search.
| Provider | How results arrive | 100,000 tweets | Rate limit | Latency |
|---|---|---|---|---|
| cheapxapi | Synchronous REST, JSON in the response | $3.00 | 2/s free, 10/s paid | Measured above |
| Official X API | Synchronous REST, pay per resource returned | $500.00 | Per-endpoint limits | Not measured by us |
| twitterapis.com | Synchronous REST, pay per call | $4.00 | Not compared | Not measured by us |
| Apify Tweet Scraper V2 | Actor run in a container, results in a dataset | $40.00 | Per Apify plan | Not measured by us |
Prices from each provider’s public pricing page; sources on the comparison page. cheapxapi pages hold up to 20 tweets or 100 follower profiles.
Re-run it yourself.
Ten profile lookups, timed from your machine. New accounts get a $1 promotional credit, so this costs nothing to try. Your numbers include the network between you and the API.
Get an API key# 10 timed requests; prints the HTTP status and total seconds of each
for i in $(seq 1 10); do
curl -s -o /dev/null -w "%{http_code} %{time_total}\n" \
"https://api.cheapxapi.com/twitter/user/info?userName=NASA" \
-H "x-api-key: $CHEAPXAPI_KEY"
doneQuestions.
Where do these numbers come from?
From the API's own request log: every paid data or account-action call made by a customer account in the last 14 days, end to end as the API saw it. Our own, monitoring and test accounts are left out. The same figures are published as JSON at /telemetry.json, refreshed when the site is built.
Why do some endpoints show no figure?
An endpoint gets a figure only when it has enough customer calls in the window. Until then the page shows how many calls there are instead of a number.
Which requests count as failures?
Server-side failures count against the success rate. Results caused by the request itself (invalid parameters, unknown user, rate limit, no balance, bad key) are left out, because they say nothing about the service.
Can I check the latency myself?
Yes. Sign up, use the $1 promotional credit, and run the script on this page: ten profile lookups cost $0.006. Your result includes your own network distance to the API.