1
test.apiqik-kratos.apiqik.com
Common issues: Extreme reasoning ability (number theory counting)、 Think supportive
94
Median
AI API relay leaderboard
pangolin aggregates public community detections and ranks relays by Bayesian-weighted score (more tests + steadier scores → higher rank). Based on 28 public reports across 7 relays (6 with ≥2 tests qualify for the main board).
🏆 Main board Top 7
Sorted by Bayesian-weighted score — more tests make the score more trustworthy
2
test.apiqik-kratos.dimleap.cn
Common issues: Extreme reasoning ability (number theory counting)、 Minimalist Input Token Audit、 Token billing
97
Median
3
unirouters.cc
Common issues: Extreme reasoning ability (number theory counting)、 Claude Tokenizer verification
96
Median
4
api.apiqik.com
Common issues: Extreme reasoning ability (number theory counting)、 Think supportive
75
Median
5
69
Median
6
27
Median
7
83
Median
How is the board scored? Click to expand the rules
Each public /r/{job_id} report feeds this board, grouped by base_url domain. Ranking uses a Bayesian-weighted score (prior=50, prior_weight=5): a one-off perfect 100 only ranks ~58.3 — you need ≥10 steady near-perfect runs to climb the main board. That stops a new relay from topping the chart after a lucky single run. The UI “Median” is the median of past scores (for intuition); sort order uses the Bayesian weight.
Single-sample relays stay off the main board (noise control) but still appear at the bottom of the full list.