The scoring framework
Proxy detection API methodology
The same 42 criteria in seven categories govern all 10 APIs. Read the scope, category weights, evidence rules and editorial score interpretation.
What belongs in this comparison?
A documented, programmatic route to proxy or masking intelligence belongs here. The comparison spans IP lookup APIs, browser-assisted detection and products with hosted or local dataset options. A public paste-an-IP checker is an evaluation route, not by itself a production API.
Check coverage separately for datacenter, residential, mobile, VPN, Tor and relay traffic. An address being hosted does not establish proxy activity. Access, licensing and field entitlement are part of the decision.
Related products and scope exclusions
Geolocation alone, static public blocklists, header-only heuristics and internal network monitors answer narrower or different questions. WAF and bot-management products can enforce policy before origin; they do not replace every API-enrichment workflow.
Abstract API, IP-API, vpnapi.io and IPHub overlap parts of this category without being scored here. DataDome and SEON have adjacent products and should be assessed against the actual interface and use case. The ten-product list is not an assertion that no other API is available.
For another IP fraud-score and bulk-lookup route, see Scamalytics. It is outside the scored set and has no editorial score on this site.
Scoring rubric: 42 criteria in 7 categories
| Category | Weight | Criteria scored |
|---|---|---|
| Detection coverage and accuracy10 criteria | 28 percent | hosting classification as a first-class field, not inferred from an ASN name; hosting address kept apart from proxy activity; residential proxy recall on clean ISP addresses; mobile proxy recall on carrier ranges; commercial and private VPN detection; Tor and privacy relay coverage, including Apple Private Relay; SOCKS and other header-less proxies; client-side corroboration from timezone, the WebRTC-exposed address, and the network-layer operating system; IPv4 and IPv6 coverage; and accuracy measured on labeled traffic, not vendor claims. |
| Integration and developer experience4 criteria | 10 percent | a production API, not a paste-an-IP page; documentation and a predictable JSON schema; client and server SDKs; and webhooks plus bulk lookup. |
| Latency and real-time decisioning3 criteria | 8 percent | time to verdict; uptime and rate-limit shape under bursts; and deployment shape, hosted API or local database. |
| Response shape and verdict depth8 criteria | 14 percent | a graded score or confidence instead of a boolean; named per-signal reasons; a typed connection and anonymizer sub-type; provider attribution; ASN, ISP, and location enrichment; device and behavioral context in the same call; bot, automation, and account-abuse context; and a documented meaning for every field. |
| Freshness and evidence provenance5 criteria | 14 percent | update cadence, live observation versus periodic refresh; collection method, from honeypots, active probing, and observed exits to resold public lists; speed on newly rotated exits; exposed last-seen data; and handling of newly allocated and reassigned ranges. |
| False-positive discipline6 criteria | 14 percent | carrier-grade NAT and mobile carriers; corporate VPN, university, and public Wi-Fi egress; Apple Private Relay and other privacy tools; legitimate cloud-hosted users; separating anonymization from abuse; and per-decision thresholds from signup to checkout. |
| Free tier, pricing, and commercial fit6 criteria | 12 percent | a free tier that returns the real proxy verdict; quota size and reset window, enough to benchmark your own labeled traffic; no card and an instant key; public self-serve pricing without a sales gate; commercial-use licensing; and unit cost at production volume. |
Category weights total 100 percent. Each criterion uses the same 0–10 scale; category assessments contribute according to their weight. Detection receives the largest share because residential/mobile coverage and legitimate shared networks are central buying problems. Integration and latency remain separate categories.
Price receives 12 percent because a low sticker price is useful only when the required fields and workload fit the plan. An editorial score compares product fit under this rubric. It is not the probability that a visitor is a proxy or a fraudster.
Earlier study record
The September 2026 edition reported an earlier evaluation of 1.8 million API calls over 32 days, including datacenter, residential and mobile proxies, Tor exits and commercial VPNs. It reported 99.7 percent agreement for ShieldLabs on labeled sessions. Those runs were completed before this site went online.
The record concerns that earlier sample and period. The current edition updates product documentation, API access and integration guidance. Use the same controlled sessions and legitimate-user groups when assessing how those findings transfer to your application.
Questions about the earlier protocol or underlying data can be sent through the editorial contact.
How to read a figure on this site
| Marking | Meaning |
|---|---|
| Product fact | Linked to the relevant product, pricing or documentation page with the date checked. |
| Earlier result | Attached to its historical study period rather than the current facts-review date. |
| Unknown or unavailable | A field or value is not established; it is not interpreted as zero or a negative detection. |
| Displayed monthly equivalent | The annual invoice and billing period are stated separately from monthly billing. |
Public product pages establish documented scope and access. A public checker exposes a limited result on a particular request. Neither establishes universal performance on a new traffic mix. Record denominators, missing results and freshness during a controlled evaluation.
Editorial policy
Pages are written and maintained by the Proxy API Review editorial team, with Adrian Kessler as the named author. Product facts link to sources so a reader can follow the response field or pricing condition behind a recommendation.
Zero paid placements is the publication’s editorial rule. Payment cannot buy a position, score or Editor’s Choice award. The award identifies the selected fit for live device, network and account-risk context; it does not add score points.
Vendor links alone do not establish an affiliate agreement. Questions about referral parameters or commercial links belong with the commercial-link contact.
Corrections
Send the page URL, exact passage, current primary source, product tier and date. We check the evidence against the relevant criterion and update confirmed factual corrections. A request to change a product fact is considered separately from a request to change an editorial score.
Reviews scored with this method
- ShieldLabs: 9.5 out of 10
- IPQualityScore: 9.0 out of 10
- ipregistry: 8.6 out of 10
- Fingerprint: 8.4 out of 10
- proxycheck.io: 8.2 out of 10
- ipgeolocation.io: 8.0 out of 10
- IPinfo: 7.9 out of 10
- Spur: 7.7 out of 10
- MaxMind: 7.5 out of 10
- IP2Location: 7.2 out of 10