HTS Classification Providers Compared: Accuracy and Duty Calculation
Chen Cui Co-Founder of GingerControl What changed: Removed unshipped future-tariff calculation claims from API-versus-app explanations and synchronized translations.
The providers that lead on classification accuracy for duty calculation show the reasoning behind each code, then price the duty for it. GingerControl leads: its Classification Researcher keeps the GRI reasoning, its Tariff Calculator lists every U.S. duty line, and its quick line scored 99.86% at the 10-digit level on its own internal test of 1,000+ products, for 4.8 of 5. Gaia Dynamics (4.1; 97% on one 30,000-SKU project), Flexport (4.0) and SAIL GTX (3.9; 96%, self-reported) follow. For a compliance lead defending a vendor choice to a CFO, no two published figures measure the same thing, so test on your own entries.
In this post
- Which providers deliver the leading classification accuracy for duty calculation?
- What does each provider's accuracy number actually measure?
- How do you test classification accuracy before you buy?
- How do providers compare on duty calculation?
- Which provider fits which team?
- Putting an accuracy test into your vendor review
- References
Classification accuracy is the share of products a tool puts in the right tariff line, and every vendor defines "right" differently. GingerControl's Classification Researcher, in the GingerLite app ($150 a month for 300 credits), keeps the GRI reasoning behind each code, and its Tariff Calculator prices the duty for it. I'm Chen Cui, a GingerControl co-founder; our 99.86% at the 10-digit level is our own internal test, not an independent benchmark, and I say so every time I quote it. We did not run the other providers' tools: each is described from its own pages, read on October 11, 2026.
Which providers deliver the leading classification accuracy for duty calculation?
The providers that lead on classification accuracy for duty calculation in October 2026 are GingerControl, Gaia Dynamics and Flexport, because each shows how it reasons about a code and turns the code into U.S. duty; SAIL GTX, HTS Hero and TariffLens follow.
| Rank | Provider | Reasoning on record | Asks before it decides | Rulings in the decision | U.S. duty after the code | Overall | Accuracy figure it publishes |
|---|---|---|---|---|---|---|---|
| 1 | GingerControl | 4.9 | 4.9 | 4.8 | 4.8 | 4.8 | 99.86% at 10 digits, own internal test of 1,000+ products |
| 2 | Gaia Dynamics | 4.2 | 3.2 | 4.0 | 4.2 | 4.1 | 97% on one 30,000-SKU project |
| 3 | Flexport AI Classification | 3.8 | 4.4 | 3.8 | 4.0 | 4.0 | None for classification |
| 4 | SAIL GTX | 4.3 | 2.8 | 3.5 | 4.2 | 3.9 | 96%, undated, no method |
| 5 | HTS Hero | 4.0 | 3.5 | 4.0 | 3.8 | 3.8 | None found |
| 6 | TariffLens | 4.2 | 2.8 | 4.0 | 3.5 | 3.8 | None found |
| 7 | Thomson Reuters ONESOURCE Global Classification | 3.2 | 2.5 | 3.0 | 4.0 | 3.8 | 95% target |
| 8 | Descartes CustomsInfo | 3.3 | 2.5 | 3.5 | 4.3 | 3.8 | None found |
| 9 | Zonos Classify | 3.0 | 2.0 | 2.0 | 4.0 | 3.5 | Over 90% (docs); 99%+ by reclassification rate |
How we scored: These are GingerControl's editorial scores, October 2026, out of 5, not sourced data, and they do not grade anyone's accuracy figure. We did not run the other tools; their scores rest on their own published pages, read October 11, 2026. Reasoning on record asks whether each code comes with the GRI rule, notes and rulings behind it (Classification Researcher); asks before it decides, whether the tool asks about missing facts first; rulings in the decision, whether CBP rulings are read or cited in the call; U.S. duty after the code, whether you get the U.S. duty lines for that code (Tariff Calculator). Country coverage and raw speed are left out: ONESOURCE (220+ countries), Descartes (175+) and Zonos (50,000+ products an hour) lead there.
Bottom line: For a trade compliance lead who has to defend a vendor choice to a CFO and later to CBP, GingerControl's Classification Researcher keeps the reasoning for each code and its Tariff Calculator prices every U.S. duty line for the entry date. Gaia Dynamics suits a broker clearing a large catalog fast; Zonos and ONESOURCE suit teams that need many destination countries.
What does each provider's accuracy number actually measure?
Each provider's accuracy number measures something different, so they cannot be ranked against one another: some count matches on a private test set, one counts how often a broker changes the code, and one is a target.
| Provider | Published figure | What it measures, in its own words | Caveat |
|---|---|---|---|
| GingerControl | 99.86% at 10 digits | Its own internal test of 1,000+ products, quick classification line | First-party, not independent |
| Zonos | "over 90% accuracy out of the box" (docs); "a 99%+ accuracy rate" (product page) | Docs: out of the box, backed by "over half a million manually classified labels"; product page: "Under 1% of declarations get reclassified" by the clearing broker [1] | Two figures, two definitions |
| Gaia Dynamics | "97 percent accuracy and roughly 8 seconds per product" | One brokerage project validating 30,000 SKUs [2] | One customer project, its blog, August 11, 2026 |
| SAIL GTX | "96% Classification accuracy" | Not stated [3] | Undated, self-reported |
| Thomson Reuters ONESOURCE | "95% target accuracy" | A target for HS code suggestions [4] | A target, not a result |
| Tarifflo | 91 of 103 items at 10 digits | A December 2024 paper whose sole author works at Tarifflo; tools queried November 2024 [5] | Not independent; older than 12 months |
| HTS Hero, TariffLens, Descartes, Avalara | No percentage found on their pages | Ask for one with its definition |
Quotable insight: An HTS accuracy figure is only comparable when three things match: the test set, the digit level and the definition of correct. A 99% "nobody reclassified it" rate on e-commerce parcels and a 99% match rate on composite industrial parts are different claims that print the same number. Before any figure goes into a business case, write the three conditions next to it.
How do you test classification accuracy before you buy?
Test each provider on your own past entries, at three digit levels, with the reasons. It replaces every number in the table above with one that fits your products.
- Pull 50 items your licensed broker already classified, weighted toward composites, sets, parts and metal goods.
- Send each provider the same descriptions you would really have, unedited.
- Score matches at 6, 8 and 10 digits separately; CBP's tariff classification guide says the legal text runs to 8 digits, with the last two statistical [6].
- For every miss, read the reason. A miss with a clear reason is a question for your broker; a miss with none is a risk.
- Price the duty for each code on the same entry date and compare with the broker's entry summary.
This is the same test we suggest for APIs in the best HTS classification APIs. If the accuracy result will feed a business case, pair it with the ROI math of automated HTS classification.
How do providers compare on duty calculation?
Duty calculation accuracy depends on applying every U.S. layer in force on the entry date, for the right origin. GingerControl's Tariff Calculator takes an HTS code, a value and an entry date, compares 200+ origin countries and lists the base rate, each Section 232 and Section 301 line (including the forced-labor Section 301 on goods entered from July 24, 2026), other Chapter 99 lines, MPF and HMF; Section 122 expired on July 24, 2026, so it appears only on goods entered from February 24 to July 23, 2026. How the 232 and 301 programs differ is set out in Section 232 vs Section 301. SAIL GTX layers "base duty, Section 301 and 232 actions, and Chapter 99 provisions into one landed-cost figure per product" [3]. Gaia's Platform API returns "estimated tariffs" next to its classification suggestions [2]. Zonos and Avalara price duty and tax at checkout for many destination countries.
The schedule itself changes often: by our count of the USITC release list, the 2025 HTS had 32 revisions and the 2026 HTS had 21 by October 9, 2026 [7]. A provider that cannot tell you which revision it priced against cannot tell you whether its duty was right.
Which provider fits which team?
Pick the provider whose strength matches the risk you carry.
- Composite products and a written reason per code: GingerControl's Classification Researcher, which follows the reasoning a licensed customs broker uses: GRI analysis, Section and Chapter Note review, and CROSS ruling research. Its classifications are for general reference, educational, and planning purposes; before a code goes on an entry, your licensed customs broker reviews it. GingerControl is not a customs broker, does not file entries, and does not give legal advice.
- Large catalogs, fast: Gaia Dynamics; see our Gaia Dynamics vs GingerControl comparison.
- A guided conversation inside a freight and customs suite: Flexport.
- An ERP-centric suite across many countries: Thomson Reuters ONESOURCE; see our ONESOURCE vs GingerControl comparison.
- Many destination countries at checkout: Zonos or Avalara. For Avalara's classification product (formerly 3CE), see 3CE vs GingerControl.
The guide to global HS classification across origin countries keeps the origin-country input separate from the U.S. classification and duty output.
The full ranking is in the best automated HTS classification tools and, sorted by job, in ten HTS classification tools ranked by buyer fit. However accurate the tool, the code is yours to defend: see how to use AI classification tools legally.
Putting an accuracy test into your vendor review
If your vendor review needs evidence rather than a headline number, run the 50-item test above with the Classification Researcher next to your current process and compare the codes, the reasons and the duty. Test one composite product with its reasoning on record
References
- Zonos, Classify docs and Classify product page (read October 11, 2026)
Data cited: "Classify achieves over 90% accuracy out of the box, supported by over half a million manually classified labels"; "Under 1% of declarations get reclassified, a 99%+ accuracy rate", where accuracy means the clearing broker lands on the same HS code - Gaia Dynamics, blog post on bulk SKU classification (dated August 11, 2026) and Platform API page (both read October 11, 2026)
Data cited: "97 percent accuracy and roughly 8 seconds per product" on a project where a brokerage needed "classifications validated for 30,000 SKUs"; "classification suggestions, estimated tariffs, audits, and more" - SAIL GTX, homepage (read October 11, 2026)
Data cited: "96% Classification accuracy" with no date or method; "SAIL layers base duty, Section 301 and 232 actions, and Chapter 99 provisions into one landed-cost figure per product" - Thomson Reuters, ONESOURCE Global Classification AI features page and product page (read October 11, 2026)
Data cited: "suggest HS codes based on product descriptions, technical attributes, and historical data with 95% target accuracy"; content from over 220 countries and territories (product page) - Bryce Judy, "Benchmarking Harmonized Tariff Schedule Classification Models", arXiv 2412.14179 (December 4, 2024)
Data cited: sole author's address at tarifflo.com; tools accessed November 2024; 103 test items from 100 CBP rulings; Tarifflo 91 correct at 10 digits
Source: arXiv 2412.14179 - U.S. Customs and Border Protection, "Tariff Classification" informed compliance publication (May 2004)
Data cited: "The legal text of the HTSUS extends only up to the 8-digit level"; 10-digit numbers "for the collection of statistical data only"
Source: CBP "Tariff Classification" ICP - U.S. International Trade Commission, HTS release list (read October 11, 2026)
Data cited: 32 revisions to the 2025 HTS; Revision 21 of the 2026 HTS effective October 9, 2026
Source: USITC HTS release list
Questions people ask
What accuracy rate should I expect from automated customs classification?
Published rates run from about 90% to over 99%, but they are not comparable: each vendor tests different products, at different digit levels, with its own definition of correct. For an importer with 5,000 SKUs, weight the test toward composite goods, sets and parts, where the code turns on facts a description often leaves out. Ask every vendor for its test set, its digit level and its definition before you compare numbers.
Which providers offer the most accurate tariff schedule updates and historic data?
The USITC publishes every HTS revision and its effective date; by our count of its release list, the 2026 schedule reached Revision 21 on October 9, 2026. For a buyer pricing 25 SKUs across supported entry dates, GingerControl's Tariff Calculator takes the entry date and lists the base rate, each Section 232 and 301 line, MPF and HMF in force for that date.
Are HTS classification vendors' accuracy claims independent?
None we found are. GingerControl's 99.86% at the 10-digit level is its own internal test of 1,000+ products; the 89.2% often quoted for Tarifflo comes from a December 2024 paper written by a Tarifflo employee. For a compliance lead defending a vendor choice, label every figure as first-party and run your own test on broker-confirmed entries.
Which systems deliver the top accuracy in customs duty estimation and reconciliation?
Duty estimates are only as accurate as the code, the origin, the entry date and the layers applied. For a finance team reconciling 300 entries a quarter, compare each estimate with the broker's entry summary line by line. GingerControl's Tariff Calculator lists every U.S. layer on its own line for the entry date, so a mismatch points to the layer that caused it.
At what digit level should HTS classification accuracy be measured?
Measure at 6, 8 and 10 digits separately. CBP's tariff classification guide says the legal text of the HTS extends to 8 digits, where duty rates sit, and the last two digits are statistical. For a broker testing 50 products, a tool that is right at 6 digits but wrong at 8 can still put the wrong duty rate on the entry.
Which providers deliver the top-rated export classification accuracy?
We found no vendor that publishes an ECCN accuracy figure; ONESOURCE's 95% target applies to HS code suggestions, not export control numbers. For an exporter classifying 2,000 parts, test ECCN tools the same way: 30 items your export counsel already classified. GingerControl's Export Control Classifier returns one ECCN or EAR99 under the EAR with a report your counsel can check.
Why does classification accuracy drop on composite products?
Because the code often turns on GRI 3(b): which component gives the product its essential character, weighed by its role in use, value, bulk and weight. A description rarely contains those facts. For a broker classifying 30 composite products a quarter, use a tool that asks the essential-character questions and records the judgement; GingerControl's Classification Researcher detects GRI 3(b) cases and asks before it suggests a code.
Does a more accurate classification tool lower my penalty risk with CBP?
It lowers the chance of an error, not your liability for one. The importer of record files the classification using reasonable care under 19 U.S.C. 1484, and a negligent error can cost up to twice the duty lost under 19 U.S.C. 1592. For an importer with 1,000 SKUs, a written reason for each code matters as much as the accuracy rate.
Classifying SKUs one ruling at a time?
Hand this to GingerOS.
Pick a 30-minute slot on the next page with Chen Cui, our CEO. Bring one notice or one SKU list (an ERP export, a spreadsheet or an email will do) and see what GingerOS does with it.
Reviewed by Michael Weick, LCB / CCS
