extractingbench

Speed and cost

How fast is it, and what does it cost, on each provider?

Each model is measured on every provider that serves it: median and 90th percentile latency, documents per minute, cold start and list-price cost per 1,000 documents.

Models we run ourselves on a Mac Mini are reported separately, with memory use and documents per minute, and never ranked against hosted providers.

Latency, median median seconds per document

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

Median wall time per document, timed from our server with the network included.

Latency, p90 90th percentile seconds per document

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

90th percentile wall time per document.

Throughput documents per minute at the stated concurrency

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

Documents finished per minute at concurrency 4, or lower where the provider's rate limit requires it.

Cost per 1,000 documents list price, US dollars

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

List-price cost per 1,000 documents, from the usage each API reports.

Test items and precision

Items
Every request in the run, per model and provider
Precision
Latency is timed from one server with the network included, and stated as such.

Full method and scoring rules