Benchmark invoice extraction with a fixed, labeled invoice set and measure both end-to-end latency and field-level accuracy. In Node.js, record request start, first streamed content when applicable, response completion, and local parsing and validation separately. There is no universally fastest API for this workload: the result depends on your invoices, routing, location, output requirements, and latency budget.
What should your latency limit cover?
Set the budget from the application requirement before comparing APIs. For an interactive workflow, decide whether the limit applies to the first visible result, the complete extraction, or both. For queued batch processing, throughput and the distribution of completion times may matter more than first-token delay.
As an Amazon Associate I earn from qualifying purchases.
End-to-end latency runs from client request initiation through response completion. It includes network travel, server processing, generation, and return transfer; a model’s generation rate is only one part. OpenAI’s latency guide identifies model choice and output-token count as important factors. There is no universal invoice-extraction threshold in the cited material, so use the limit your application actually needs.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →How do you build a fair invoice benchmark?
Fix the corpus and expected outputs
Use the same representative invoices for every candidate configuration. Preserve the source documents and include the range of layouts and document lengths your application expects. Label each invoice with the target fields and specify how line items should be represented if they are in scope. The corpus and schema are part of the benchmark definition, not details to change between providers.
#1 Best Overall
- PORTABLE SCANNER FOR USE ON-THE-GO — The fastest and lightest mobile single-sheet-fed compact document scanner in its class¹
- QUICK DOCUMENT SCANNING ― This Epson ultra-fast scanner scans a single page as quickly as 5.5 seconds²; Windows and Mac compatible
- VERSATILE PAPER HANDLING ― Portable scanner scans documents up to 8.5 x 72 in; Also easily digitizes receipts and ID cards to make accounting, bookkeeping, and organizing simpler
- INTUITIVE, HIGH-SPEED SOFTWARE — Epson ScanSmart Software³ is a smart tool allowing you to easily scan, review, and save; Stay organized easily with the help of this Epson scanner
- EASY SETUP — USB-powered connect to your computer for quick and simple scanning; No batteries or external power supply required to operate portable document scanner; Standard Connectivity: USB 2.0
Hold inputs and conditions constant
Keep the prompt, target fields, output schema, generation limits, concurrency pattern, client location or region, and Node.js and SDK setup as consistent as provider capabilities allow. Document unavoidable differences, such as inline PDF input versus upload-and-reuse or preprocessing to text. If timing begins after a file has been uploaded, say so: that measures a different path from timing that includes upload.
Google’s document-processing guide supports PDF input and structured extraction. It recommends its Files API for large or reused files, where decoupling upload from model requests can reduce request latency and bandwidth. This is guidance for Google’s service, not a rule that applies to every provider.
Separate warm-up from measured runs
Use a defined warm-up phase and report it separately from steady-state measurements. OpenAI says a first API response using a new Structured Outputs schema may require additional processing, with later uses cached. Its 2024 announcement reported that typical schemas took under 10 seconds to process on first use and more complex schemas could take up to a minute; those are announcement figures, not a prediction for every current schema. Do not silently discard slow first requests: state whether they are included, excluded, or reported separately. Check current model availability and limitations in the Structured Outputs documentation.
Rank #2
- FAST SPEEDS - Scans color and black and white documents a blazing speed up to 16ppm (1). Color scanning won’t slow you down as the color scan speed is the same as the black and white scan speed.
- ULTRA COMPACT – At less than 1 foot in length and only about 1. 5lbs in weight you can fit this device virtually anywhere (a bag, a purse, even a pocket).
- READY WHENEVER YOU ARE – The DS-640 mobile scanner is powered via an included micro USB 3. 0 cable allowing you to use it even where there is no outlet available. Plug it into you PC or laptop and you are ready to scan.
- WORKS YOUR WAY – Use the Brother free iPrint&Scan desktop app for scanning to multiple “Scan-to” destinations like PC, Network, cloud services, Email and OCR. (2) Supports Windows, Mac and Linux and TWAIN/WIA for PC/ICA for Mac/SANE drivers. (3)
- OPTIMIZE IMAGES AND TEXT – Automatic color detection/adjustment, image rotation (PC only), bleed through prevention/background removal, text enhancement, color drop to enhance scans. Software suite includes document management and OCR software. (4)
How do you measure LLM API latency in Node.js?
Use a monotonic high-resolution timer for elapsed durations. Node.js v24.21.0 documents performance.now() in node:perf_hooks as a millisecond timestamp relative to process start. Use the runtime deployed by your application and consult the Node.js performance hooks documentation.
import { performance } from 'node:perf_hooks';
const started = performance.now();
let firstContentMs;
// In the provider SDK's stream loop, record the first non-empty content event:
// firstContentMs ??= performance.now() - started;
// After the response is complete and local parsing/validation finishes:
const completedMs = performance.now() - started;This illustrates timing boundaries, not a provider-specific stream implementation. SDK event names and response properties differ, so use the relevant provider SDK documentation rather than assuming one universal event name.
For every request, save an individual record in JSONL or CSV with at least these fields:
Rank #3
- Fast and Efficient: Scans both sides of a document at the same time, in color, at up to 45 pages per minute, with a 60 sheet automatic feeder, and one touch operation. Innovative Feeding System.
- Reliably Handles Many Different Document Types: Receipts, business cards, reports, contracts, long documents, thick or thin documents, and more. Monochrome LCD Display.
- Designed exclusively for the included Canon CaptureOnTouch software;TWAIN and ISIS drivers are not supported.
- Easy Setup: Simply connect to your computer using the supplied USB-C cable.
- Bundled Software: Includes easy-to-use Canon CaptureOnTouch scanning software.
- Request start, first non-empty streamed content time when available, response completion, and local parse-and-validation completion.
- Outcome, error or timeout, and retry information.
- Provider, model, and actual route metadata returned by the API or gateway, when available.
- Input and output token counts when provided, plus the timing interval used for any token-rate calculation.
- Whether upload or preprocessing time is included and whether the request was warm-up or a measured run.
Call a measurement time to first token (TTFT) only when the response is streamed and the first-token boundary is measured consistently. For non-streaming calls, report request-to-completion latency instead. Google Cloud’s inference metrics documentation distinguishes TTFT, time per output token (TPOT), inter-token latency (ITL), request latency, and throughput. It defines TPOT as (request_latency - time_to_first_token) / (total_output_tokens - 1) and NTPOT as request_latency / total_output_tokens. It also cautions that requests per second can vary with context length. These metric definitions are useful for interpretation, but are not an SLA or a hosted API benchmark.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →What should you report besides average completion time?
For repeated runs, publish p50 and p95 latency, and p99 if the sample size supports it. Include successful-request and failure counts. For streamed workloads, show TTFT separately from completion latency; do not combine the two into one number. Report token counts and tokens per second only with a clear numerator and timing window.
Compare the returned values with the labeled outputs. Choose and explain field-level metrics—for example, exact match after a declared normalization rule—and include schema-valid rate, missing-field rate, incorrect-field rate, and a task-specific line-item measure when relevant. Structured output can enforce a response shape, but it does not establish that the values are correct. OpenAI states that “Structured Outputs doesn’t prevent all kinds of model mistakes” in its Structured Outputs announcement.
Rank #4
- OUR MOST ADVANCED SCANSNAP. Large touchscreen, fast 45ppm double-sided scanning, 100-sheet document feeder, Wi-Fi and USB connectivity, automatic optimizations, and support for cloud services. Upgraded replacement for the discontinued iX1600
- CUSTOMIZABLE. SHARABLE. Select personalized profiles from the touchscreen. Send to PC, Mac, mobile devices, and clouds. QUICK MENU lets you quickly scan-drag-drop to your favorite computer apps
- STABLE WIRELESS OR USB CONNECTION. Built-in Wi-Fi 6 for the fastest and most secure scanning. Connect to smart devices or cloud services without a computer. USB-C connection also available
- PHOTO AND DOCUMENT ORGANIZATION MADE EFFORTLESS. Easily manage, edit, and use scanned data from documents, receipts, photos, and business cards. Automatically optimize, name, and sort files
- AVOIDS PAPER JAMS AND DAMAGE. Features a brake roller system to feed paper smoothly, a multi-feed sensor that detects pages stuck together, and skew detection to prevent paper damage and data loss
Include cost per invoice or another clearly defined cost basis using provider prices checked at publication time. The cited material does not establish a stable, comparable price table for this benchmark.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How do PDF handling and gateway routing affect the result?
Make the measured document path explicit
PDF upload can sit inside or outside the timed interval. Report whether each request includes file transfer, reuses an earlier upload, or sends preprocessed text. Google recommends the Files API for larger or reused PDFs because separating upload from inference can reduce request latency and bandwidth on its service; that can make a per-inference timing incomparable with a test that includes upload.
Record routing and fallback behavior
A unified gateway can introduce routing choices, provider fallback, cold caches, or account checks. Record routing preferences, fallback configuration, and actual provider or route metadata whenever exposed. OpenRouter describes possible latency contributions from cold edge caches, billing checks, and fallback attempts, and documents provider preferences for latency goals in its provider routing documentation. These are statements about OpenRouter’s behavior, not a measured fixed overhead for all gateways.
Best Value
- FAST DOCUMENT SCANNING — Document scanner with feeder allows you to speed through stacks with a 50-sheet Auto Document Feeder (ADF); Efficient office scanner to help you scan more productively
- INTUITIVE, HIGH-SPEED SOFTWARE — Quickly scan with this desktop document scanner; Epson ScanSmart Software lets you easily preview scans, email files, upload to the cloud, and more; Plus, automatic file naming saves even more time
- SEAMLESS INTEGRATION — Easily incorporate your data into most document management software with the included TWAIN driver; Office document scanner integrates seamlessly with business workflows
- EASY SHARING — Duplex scanner allows you to scan straight to email or popular cloud storage2 services like Dropbox, Evernote, Google Drive, and OneDrive for simple storage and sharing
- SIMPLE FILE MANAGEMENT — Scanner allows the creation of searchable PDFs with Optical Character Recognition (OCR) and convert scans to editable Word or Excel files effortlessly; Designed for home and office document scanning
Which LLM API is fastest for invoice extraction?
That cannot be determined from feature documentation alone. A credible answer requires timing the same invoice corpus, schema, client location, concurrency, and file path against the specific API and model configurations under consideration. Compare the candidates across the dimensions that affect your deployment:
- Tail latency: p50 and p95 completion time, plus TTFT when users see streamed output.
- Extraction quality: field correctness against labels, schema-valid rate, missing values, and line-item performance if applicable.
- Document fit: PDF support, structured-output support, file reuse, size limits, and preprocessing requirements.
- Routing: fixed versus automatic provider selection, fallback policy, and visibility into the route actually used.
- Operations and cost: cost on the same workload, errors, timeouts, rate limits, and retry policy.
Do not collapse these into a single winner score unless you explain how your application weights speed, correctness, reliability, and cost.
Which changes are worth testing?
OpenAI’s latency guidance suggests reducing generated tokens, using smaller structured-output syntax, avoiding sequential requests where possible, and parallelizing independent subtasks. It gives a general heuristic that removing 50% of output tokens may cut latency by roughly 50%; it also says cutting 50% of a prompt may improve latency by only 1–5% for ordinary requests, with massive contexts potentially behaving differently. These are general vendor observations, not guarantees for invoice extraction.
Test each change on the same corpus and compare both latency and field correctness. Shorter outputs or prompts are not improvements if they omit required values; parallel work is appropriate only when subtasks are independent. For bulk work with no strict interactive deadline, Google’s Gemini Batch API is asynchronous and lists a 24-hour target turnaround in its page last updated 2026-09-01 UTC; it is not a fit for a strict interactive latency limit. See Gemini Batch API documentation.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
