DATA SHEET · 2026-09-22 · READ TIME 3 MIN
Benchmarks: measured, not estimated
Every number on this page comes from a real run. Test machine: Windows 11 Home, Intel Core i3-1115G4 (2 cores, 4 threads, 3.00GHz), 8GB RAM, one of the weakest laptops you can still buy. If jev-seo is fast here, it is fast on your hardware and in CI.
The numbers
| Metric | Value | How measured |
|---|---|---|
| Release binary | 6.7MB (6,996,992 bytes) | v0.1.0 Windows release asset, file size |
| Download zip | 3.2MB (3,215,386 bytes) | v0.1.0 Windows release asset, file size |
| RAM footprint | Under 30MB | Designed ceiling, Rust single binary |
| Audit, 6-file site | Seconds, one pass | Reading structure, schema, headings, copy |
| GEO direct answer | 0.90 confidence | geo command on index.html |
| GEO composite | 7/10 | After direct-answer and statistics fixes |
| SERP scrape | No API key | Free DuckDuckGo endpoints |
| Search visibility | #1 for own term | query command, "jev seo" |
| Jev inference | 70 to 500ms | TypeSafe Jev API spec |
| Jev throughput | 250k tok/s, 1200 req/min | TypeSafe Jev API spec |
| Jev cost | ~$0.0004 per decision | List price, output tokens free |
What the audit pass actually checks
One jev-seo audit . run on this site's local files reported: total words before fixes 1,651, after fixes 2,655; overall pass rate before 53.1 percent, after 72.2 percent. It caught four orphan pages with zero incoming links, four pages under the word threshold, one missing canonical, and one duplicate content source, each with the file and rule named. The fixes took one editing session, not a dashboard subscription.
GEO scoring, the benchmark nobody else publishes
The geo command scored the homepage for the query "jev seo" and reported a direct answer with 0.90 confidence. Before the copy rewrite, the same page scored around 0.35 with a "no direct answer" verdict. The delta is the difference between being a page on the internet and being a page a generative engine can quote.
Why these numbers matter
Hosted SEO tools bill $130 a month and still cannot tell you whether an AI engine would cite your page. jev-seo runs the whole audit, GEO, robots, sitemap, and rank loop locally, in under 30MB RAM, on hardware from 2020, for zero subscription. When an agency reports "$250 per client audit" in the current Jev-for-SEO wave, this is the tool that makes the same workflow cost fractions of a cent.
Frequently asked questions
Are the transfer speeds and crawl times included? No. These numbers are local file operations and API calls. Network crawl speed depends on your connection and the target site.
Does RAM stay under 30MB during a big audit? The ceiling holds for the local audit path. The rank history SQLite file is local and small; a large site with thousands of pages grows storage, not memory.
How do I reproduce these numbers? Install v0.1.0, clone the repo, run jev-seo audit . on the site branch, then jev-seo geo --query "jev seo" --target index.html with a free TypeSafe key.
NEXT: comparisons → · guide → · home