Ranking methodology

Current status: two ranking signals, one still in developmentRankings are currently led by the Hivig Velocity Index — a 0–100 score drawing on real OpenRouter token volume, Hugging Face downloads, and LiveBench quality data — for any model an editor has opted into automated scoring. Models without that opt-in fall back to sorting by release date (newest first) until they're added. A second score, the Hivig Score, is in development — it will assess capability, agentic performance, and economics using Hugging Face, Epoch AI, and Berkeley Function-Calling Leaderboard (BFCL) data. As with the Velocity Index, the exact formula and category weights won't be published once it ships.

How this methodology evolves

Ranking AI models well is an ongoing process. We're actively refining the Hivig Score and Velocity Index against real-world agentic use cases — this isn't a finished formula, and it shouldn't be treated as one.
The methodology itself is subject to change as a result, and we'll keep this page current as it does.
We're committed to giving our users transparent information. The core guiding principles behind how we rank models will always be published here — even in cases where the exact formula stays undisclosed.

Full detail lives in RANKING_METHODOLOGY.md in the project source.