Data Extraction Race — results
Invoice — Blue Harbor Logistics
$0.0076 total
Scoreboard
#1 Claude Opus 4.7Anthropic★ Won this round on this document
94/100
1940 ms · $0.0051
| Field | Expected | Model answer | |
|---|---|---|---|
| tax | 256 | 256 | exact |
| total | 3456 | 3456 | exact |
| vendor | Blue Harbor Logistics | Blue Harbor Logistics | exact |
| customer | Greenfield Markets | Greenfield Markets | exact |
| due_date | 2026-06-11 | 06/11/2026 | close |
| subtotal | 3200 | 3200 | exact |
| invoice_date | 2026-05-12 | 05/12/2026 | close |
| account_last4 | 8842 | 8842 | exact |
| payment_terms | Net 30 | Net 30 | exact |
| invoice_number | INV-2026-0451 | INV-2026-0451 | exact |
#2 Claude Sonnet 4.6Anthropic
85/100
2535 ms · $0.0025
| Field | Expected | Model answer | |
|---|---|---|---|
| tax | 256 | $256.00 | close |
| total | 3456 | $3,456.00 | close |
| vendor | Blue Harbor Logistics | Blue Harbor Logistics | exact |
| customer | Greenfield Markets | Greenfield Markets | exact |
| due_date | 2026-06-11 | 06/11/2026 | close |
| subtotal | 3200 | $3,200.00 | close |
| invoice_date | 2026-05-12 | 05/12/2026 | close |
| account_last4 | 8842 | 8842 | exact |
| payment_terms | Net 30 | Net 30 | exact |
| invoice_number | INV-2026-0451 | INV-2026-0451 | exact |
Source document
INVOICE #INV-2026-0451 from Blue Harbor Logistics to customer Greenfield Markets. Invoice date: 05/12/2026. Payment due: 06/11/2026. Line items: freight handling, customs clearance, and last-mile delivery. Subtotal: $3,200.00. Tax (8%): $256.00. Total due: $3,456.00. Payment terms are Net 30. Please remit to account ending 8842. Extract the following fields as JSON: invoice_number, vendor, customer, invoice_date, due_date, subtotal, tax, total, payment_terms, account_last4.
How scoring works
Every model received the identical prompt and field list and answered once at temperature 0. We compare each field to the ground truth: an exact match earns full credit, the same value in a different format earns partial credit, anything else earns none. No model judges another — the score is pure field matching. A win means this model scored highest on THIS document with THESE fields, not that it is better overall.