I buy things from foreign websites and haveEcommerce & Shopping
How each model performed on this task. Every chart states its N, and the reward carries a 95% confidence interval once there is more than one run to measure.
Core metrics
Check Pass %
Mean share of the task's checks passing, averaged over runs. N = 1 run per model.
85%
Claude Opus 5.5
54%
Claude Opus 5
46%
GPT-5.6 sol
38%
GPT-6 sol
Extended metrics
Median Time
Wall-clock minutes, median across runs. N = 1 run per model.
7m
Claude Opus 5
3m
GPT-5.6 sol
Tool Calls
Tool invocations, median across runs — counted from the trajectories. N = 1 run per model.
53
Claude Opus 5.5
40
GPT-6 sol
38
Claude Opus 5
15
GPT-5.6 sol
Verifier × model pass rate
How often each verifier passed, per model. Green is passing, red is failing — a red band across a row is a verifier nothing clears.
| Verifier | Claude Opus 5.5n=1 | GPT-6 soln=1 | GPT-5.6 soln=1 | Claude Opus 5n=1 |
|---|---|---|---|---|
| checks (13) | ||||
| 100% | 100% | 100% | 100% | |
How to passOpen https://www.customs.go.kr/english/cm/cntnts/cntntsView.do?mi=8067&cntntsId=2741 and record the three transaction types described for overseas purchases and how they differ. Then record the rule that determines which clearance procedure applies, naming all three carriers or channels it distinguishes. This rule is what the rest of the task turns on. | ||||
| 100% | 100% | 100% | 0% | |
How to passOpen https://www.customs.go.kr/english/cm/cntnts/cntntsView.do?mi=8057&cntntsId=2733 and record the three types of declaration with the value band for each, including the higher band that applies to goods from one specific country, and the value above which the general declaration is required. Record what an item list declaration consists of and what identifier of the owner it requires. | ||||
| 100% | 100% | 100% | 0% | |
How to passFrom the same page, list all the categories of goods excluded from the item list declaration. State what happens to an excluded item even when its value is below the threshold, and name three everyday online purchases that would be caught by this list. | ||||
| 100% | 0% | 100% | 0% | |
How to passFrom the same page, record the exemption rule for personal-use goods, which taxes are still charged despite an exemption, and how the price of goods is defined including what may be excluded and on what condition. The main sentence and the footnote treat international shipping differently; quote both and say plainly that they do not read the same. | ||||
| 100% | 100% | 100% | 100% | |
How to passOpen https://www.customs.go.kr/english/cm/cntnts/cntntsView.do?mi=8058&cntntsId=2734 and record the value below which postal items are exempt, the two declaration types, and every trigger that forces a general import declaration, including the value threshold and the separate gift value in local currency. Record who must lodge that declaration. | ||||
| 100% | 0% | 0% | 100% | |
How to passUsing steps 2 and 5 only, state for an order worth USD 900 and again for an order worth USD 1,500 which declaration type applies if it comes by courier and which if it comes by post. Name the threshold that differs between the two channels and by how much. State which channel forces a broker sooner. | ||||
| 100% | 0% | 0% | 0% | |
How to passOpen https://www.customs.go.kr/english/cm/cntnts/cntntsView.do?mi=8028&cntntsId=2716 and record the general duty-free allowance for a traveller's personal effects, the three additional allowances with their quantity limits, the age restriction, and the three conditions attached to agricultural, fishery and livestock products including the weight ceiling and the acquisition price limit. | ||||
| 100% | 0% | 0% | 100% | |
How to passTake one order of goods worth USD 500 for personal use. State whether duty and VAT would apply if it arrived by courier, by post, and carried in your suitcase, citing the specific threshold used in each case. Then state which of the three treats that value most favourably, and by what margin. | ||||
| 100% | 100% | 100% | 100% | |
How to passOpen https://www.customs.go.kr/english/cm/cntnts/cntntsView.do?mi=10801&cntntsId=5502 and record why the code system exists, what it replaces, and exactly what a foreigner must enter when importing e-commerce goods, including which two identifiers are explicitly not accepted. Tie this back to the identifier requirement you recorded in step 2. | ||||
| 0% | 0% | 0% | 0% | |
How to passFollow the issuance link from the page in step 9. If the UNI-PASS page is publicly reachable, record which language it renders in and whether an English path is offered, without registering, authenticating, or entering personal information. If the page cannot be reached or rendered, record that limitation instead and do not infer its language or English availability. State what this means for someone who does not read the displayed language. | ||||
| 0% | 0% | 0% | 0% | |
How to passFollow the issuance link from the page in step 9. If the UNI-PASS page is publicly reachable, record which language it renders in and whether an English path is offered, without registering, authenticating, or entering personal information. If the page cannot be reached or rendered, record that limitation instead and do not infer its language or English availability. State what this means for someone who does not read the displayed language. | ||||
| 100% | 0% | 0% | 100% | |
How to passState whether these English pages address orders split across several parcels arriving together, whether the exemption is per parcel or per person over a period, and whether they give a duty rate for goods above the threshold. Where they do not, say so plainly and name the page you checked. Do not supply a figure from any other source. | ||||
| 100% | 0% | 0% | 100% | |
How to passProduce a ranked list of the three channels for a single personal-use order, best first. For each, give the value ceiling that keeps it simple, the paperwork it triggers above that, the identifier or document needed, and the specific risk that would push the order into the next tier. Close with one recommendation and the sentence from the sources that decides it. | ||||
Failure modes on this task
23 tagged verifier failures, by the stage they were assigned to. Higher is worse.
| Failure stage | Claude Opus 5.5 | GPT-6 sol | GPT-5.6 sol | Claude Opus 5 |
|---|---|---|---|---|
| Delivering the result1 mode | — | 2 | 3 | — |
| Standing up the evidence1 mode | — | 2 | 1 | 1 |
| No signal in the judge’s words1 mode · the grader | 2 | 4 | 3 | 5 |