For Claude Code, Cursor, Codex, opencode, Pi and Hermes. One install, every agent.
01 / SuperSearch
What SuperSearch is, defined →The full benchmark, every call public →
02 / Observability
Swipe for the rest of the night · tap a mark to read the turn
6 of 6 engines answered54 or fewertimeout or quota in the turnan engine first named the winnerAP call
Six of 156 election-night sessions. See them all, turn by turn →
03 / Setup
$ curl -fsSL https://jevlin.ai/install.sh | sh
npm install -g @jevlinsearch/jevlin && jevlin setupirm https://jevlin.ai/install.ps1 | iexand supports
04 / Pricing
A search is billed at the cost of the indexes it asks, with no markup. The tier sets how much work one query does.
Sign-up credit: $10.
Weekly credit: $2.
Pages: $0.002 each. Cached: free.
Trajectories: free, kept 30 days.
Enterprise: enterprise@jevlin.ai
A routine lookup. Two indexes by default, three on standard.
More than one source, or a thin first answer. 8 results each.
Corroboration, with longer excerpts.
05 / Benchmark
No model judge where the answer page is known. Every call is published, with the data and the scripts beside each result.
3 Nov 2026
Election night: which engine inside the fan-out finds each called race first?
156 races timed, live, one call each; the ranking of engines by who named the winner first.
29 Oct 2026
Can the fast tier beat the best single engine at a fraction of its price?
Economy, $0.0013: 60%. Standard, $0.0053: 62%. Exa, $0.007: 58%.
27 Oct 2026
One index or the fan-out, on 485 developer questions?
Balanced 62%, best single engine 58%. Three phrasings lift every engine by 12 to 16 points.
27 Oct 2026
Do search APIs agree? 4,365 queries to four engines.
Two engines share 2.1 to 2.5 pages of 10. A page all four return is the answer 25% of the time; a page one returns, 2.6%.