Invoice and receipt extraction reads a supplier document and returns the fields an accounts system needs: supplier name and tax number, invoice number and date, each line item with quantity and price, net, tax and total.
Two properties make this different from ordinary extraction. First, the document must be internally consistent — line items should sum to the net, and net plus tax to the total — which gives a free accuracy check no other extraction task has. Second, most organisations receive documents from a few hundred recurring suppliers, so a model fitted to those layouts outperforms a general one. A working pipeline therefore combines extraction, arithmetic validation, and a review queue for anything that fails the sums.
Models in this group take text or whole documents as input: plain text, PDFs, scanned pages and office files. Each specification table gives three hardware tiers — Minimum, the smallest setup on which the model runs correctly; Medium, the usual production configuration; and High, a configuration sized for peak volume. The rate is the number of sample inputs processed per hour on that hardware. Use these rates for initial sizing. Before production, benchmark your own data to validate accuracy, latency, throughput and cost. The sample input here is one A4 invoice at 300 dpi with header fields and line items extracted.
LayoutLMv3
Vendor: Microsoft
What it does: reads the words on an invoice together with their positions, so it distinguishes the total from a line amount by where it sits. Fitted to your recurring supplier layouts, it is the most accurate and cheapest option per document.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (documents/hour)
≈ 1,600
≈ 6,000
≈ 15,000
Donut
Vendor: NAVER Clova
What it does: extracts the fields straight from the page image with no character-reading step, which removes a source of error on faxed and poorly photocopied invoices.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (documents/hour)
≈ 900
≈ 3,200
≈ 8,000
PP-StructureV3
Vendor: Baidu (PaddlePaddle)
What it does: reads layout, tables and text in one pass, which recovers invoice line-item tables cleanly. A sound default when line detail matters as much as the header fields.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
L40S 48 GB
VRAM
8 GB
24 GB
48 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× L40S 48 GB
Rate (documents/hour)
≈ 900
≈ 3,600
≈ 6,500
Surya
Vendor: Datalab
What it does: reads invoices in roughly ninety languages, which matters for cross-border payables where documents arrive in the supplier’s own language and script.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (documents/hour)
≈ 700
≈ 3,000
≈ 6,000
Input type — Pictures
Models in this group take a photograph as input: a receipt or invoice photographed by a member of staff on a phone. The three hardware tiers mean the same as above. The sample input here is one photograph of a receipt at about 4 megapixels.
Qwen2.5-VL 7B / 72B
Vendor: Alibaba Cloud
What it does: reads a photographed receipt — creased, curved, poorly lit — and returns the fields as a structured record, following your own field list. The most robust option for expense claims captured on phones.
Requirement
Minimum
Medium
High
GPU type
RTX 4090 (7B, reduced precision)
A100 80 GB (7B, full precision)
2× H100 80 GB (72B model)
VRAM
16 GB
80 GB
160 GB combined
vCPUs
8
16
32
RAM
32 GB
64 GB
200 GB
Server
1× RTX 4090 24 GB
1× A100 SXM 80 GB
2× H100 SXM 80 GB
Rate (images/hour)
≈ 400
≈ 1,600
≈ 900 (72B model, higher accuracy)
InternVL 3
Vendor: OpenGVLab (Shanghai AI Laboratory)
What it does: the same task with particular strength on dense line-item tables, such as long supermarket or hardware receipts.
Requirement
Minimum
Medium
High
GPU type
RTX 4090 (2B model)
A100 80 GB (8B model)
4× H100 80 GB (38B model or larger)
VRAM
12 GB
40 GB
320 GB combined
vCPUs
8
16
48
RAM
32 GB
64 GB
256 GB
Server
1× RTX 4090 24 GB
1× A100 SXM 80 GB
4× H100 SXM 80 GB
Rate (images/hour)
≈ 550
≈ 1,800
≈ 1,200
Florence-2
Vendor: Microsoft
What it does: a small vision model that locates and reads the key regions of a receipt. Fast and cheap enough to run on every photograph as a first pass, escalating only unclear ones.
Requirement
Minimum
Medium
High
GPU type
RTX 3060
RTX 4090
A100 80 GB
VRAM
8 GB
24 GB
80 GB
vCPUs
6
12
24
RAM
16 GB
32 GB
64 GB
Server
1× RTX 3060 12 GB
1× RTX 4090 24 GB
1× A100 SXM 80 GB
Rate (images/hour)
≈ 1,800
≈ 6,000
≈ 14,000
Choosing between them
The choice depends on how many suppliers you receive from, how varied their layouts are, and whether documents arrive as PDFs or as photographs. Our consultants review a sample of your intake and recommend a model, the validation rules, and the confidence threshold above which an invoice can post without review.
At the start of a project we may run a short proof of concept on a sample of your own data, measuring the accuracy and the throughput the model actually achieves on your material. That replaces the estimates on this page with real figures, so the cost and the schedule for the full engagement are known before it is committed.
Share a representative sample, expected volume, latency target and deployment location for Invoice / receipt extraction. We benchmark the shortlisted models, recommend the lowest-cost GPU configuration that meets the target, and scale it from pilot capacity to a dedicated production cluster.