Source transparency
AI benchmark data sources
See what every source measures, how Vibe Bench uses it, when its data was last observed, and what it cannot prove.
Artificial Analysis
Vibe Bench reads the provider scores through the documented OpenRouter benchmark feed and keeps each source value before any blend.
Arena Code WebDev
Vibe Bench matches a public Arena model name conservatively, normalizes its ELO to a fixed scale, and blends it into Coding only while the result is current.
Design Arena
Vibe Bench stores category results as a separate evidence family. These rankings do not change the main Vibe Coding Index.
How Vibe Bench protects the rankings
Every source keeps its native value, date, status, and link before normalization. Preliminary results remain labeled, stale evidence loses confidence, and focused category rankings stay separate from the main VCI until the public methodology changes.
Read the complete methodology →