Royal Gold Cursor
Knight AI+AV Mark KNIGHT AI+AV
Home
Rankings
Start a project
Home / Rankings / AI Rankings
Knight Rating · 6 Sources · Every Hour

AI Rankings.
One Knight Rating. Six boards.

Live mirror of Artificial Analysis. The Knight Rating is their intelligence index normalized to 0–100; tied scores keep AA's native order so the table stays in lockstep with their site. Coding, speed, and cost are shown for context, not weighted. Pulled every hour.

6
Source boards
1h
Cadence
—
Models tracked
—
Last refresh
# Model Knight Rating Intel. Speed Cost Coding Δ
Loading Knight Rating board…
Source Boards · Weighted Knight Rating

The six we trust.

Each board is pulled fresh, normalized, and weighted into the Knight Rating. Weights are tunable per use-case — below is our current default.

Loading sources…
How this runs

The Knight Rating method.

  1. 01

    Pull

    Knight Workers fetch fresh boards from all six sources every hour.

  2. 02

    Normalize

    Each source is rescaled to a 0–100 band so a Chatbot Arena Elo and a SWE-bench pass-rate are directly comparable.

  3. 03

    Reconcile

    Our on-board AI matches aliased model names (e.g. "claude-opus-4-7", "Claude Opus 4.7") into a single canonical record.

  4. 04

    Weight

    Knight Rating = 0.70·intelligence + 0.20·coding + 0.07·speed + 0.03·cost. Axis scores are the Σ (source_score × source_weight) across the six boards. All weights live in rankings.json → methodology and are editable without code changes.

  5. 05

    Publish

    Fresh rankings.json is signed and written to the site. Completely isolated from the Knight AI+AV production database.