Measured performance
Devana’s document extraction engine reads a 500-page PDF in 1.5 s, keeps your tables intact and saves the GPU for the pages that need it. Entirely within your infrastructure.
Measured on November 19, 2026 · ODIN 2.0.32, extraction engine 0.2.0 · one container, 4 CPUs
Speed
ODIN returns the complete content of a 500-page PDF in 1.5 s. Four CPUs are enough.
Up to1,000×faster than the systems compared.
Page readPage waitingIn real time, at the measured speed: one square per page.
PDF · 500 pages
1.5 s
341 pages per second
23× to 1,000× faster
Word · 500 pages
0.04 s
11,900 pages per second
3.5× to 160× faster
PowerPoint · 100 slides
0.01 s
14,300 slides per second
7.8× to 110× faster
Engine times in fast mode, median of 5 runs. End to end through the service, queue and storage included: 2.3 s for the PDF, 0.86 s for Word, 0.38 s for PowerPoint. 9 systems compared across these three formats, with their default settings.
Tables
In a price list, the word “Included” covers three plans. ODIN keeps that span: it renders your tables in HTML, merged cells included.
In Markdown
| Service | Plan A | Plan B | Plan C |
|---|---|---|---|
| Support | Included |
| Support | Included | | |With ODIN, in HTML
| Service | Plan A | Plan B | Plan C |
|---|---|---|---|
| Support | Included | ||
<td colspan="3">Included</td>8 of 8
ODIN, from Word
7 of 8
ODIN, from PDF
0 of 8
Systems compared, default output, from PDF
Test document: three tables, 42 cells, 8 of them merged. A merge is counted when the cell appears in the output with its exact span. All 42 cells appear in ODIN’s output.
Fidelity
On a digital document, ODIN reads the text as it is written in the file. On a scanned page, a vision model hosted on your servers takes over.
The vision model is a setting: you replace it through configuration.
40 real pages of public documents, compared word for word with their reference text; medians over the 18 pages read by the native parser. Across all 40 pages, with images and tables transcribed by the vision model: 92.0% fidelity. Mistral Small 3.2 and Mistral Medium 3.5: general-purpose vision models, deployable on-premises.
Energy
ODIN chooses the path page by page: text is read on CPU, only image pages go to the model on GPU.
2,500×less energy, at minimum, for a page read on CPU.
half a secondwith ODIN: 5.2 J
2 h 20 min to 75 hwith an inference on every page: 84 kJ to 2.7 MJ
Word, 500 pages
99.99%
of energy saved
500 pages on CPU · 0 on GPU
5.2 J with ODIN, versus 84 kJ to 2.7 MJ by inference
PowerPoint, 100 slides
99.99%
of energy saved
100 slides on CPU · 0 on GPU
1.2 J with ODIN, versus 17 kJ to 536 kJ by inference
PDF, 500 pages, fast mode
99.97%
of energy saved
500 pages on CPU · 0 on GPU
27 J with ODIN, versus 84 kJ to 2.7 MJ by inference
PDF, 500 pages, default mode
61%
of energy saved
307 pages on CPU · 193 on GPU
32 kJ to 1 MJ with ODIN, versus 84 kJ to 2.7 MJ by inference
Default mode: 193 of 500 pages contain images or tables sent to the model.
Architecture
Each page takes the path that suits it. Text is read straight from the file, on CPU; image pages are transcribed by a vision model hosted on your servers. The whole process runs within your infrastructure.
Document
PDF, Word, PowerPoint, image
Page analysis
Page-by-page routing
The page carries its text
Native path
Parser · CPU
The page is an image
Vision path
On-premises model · GPU
Structured content
Page by page, tables in HTML, route of each page
Deployment
The native path runs on standard CPUs. The speed measurements on this page were taken on 4 CPUs.
For image pages: hosted on your servers and replaceable by configuration.
The native path runs locally; the vision path calls the model hosted on-premises.
Whatever the document: structured content page by page, tables in HTML, route of each page.
The GPU is called on for image pages only.
366 catalogue
These applications of the catalogue rely on ODIN to read your documents.
By default, ODIN processes every file dropped in: your AIs access its content directly.
View the appYou define the fields to extract; IDP relies on ODIN to read the documents received.
View the appODIN extracts the content of your on-premises SharePoint sites, indexed for your agents.
View the appOur teams present ODIN and its deployment within your infrastructure.