SIGNALAF
◈ Guide

AI Rating

Rating the model, the tool, or the user are three different jobs.

Short answer: an AI rating scores one of three things. Model ratings say which model is more capable on average, tool ratings say which product fits a job, and operator ratings say how efficiently a person drives AI. They answer different questions and should not be mixed.

Last updated 2026-11-06.

Three kinds of AI rating

RatesHowAnswersLimitMore
AI modelsStandardised test suites and human-preference voting. Arena-style sites rank models by blind side-by-side votes with Elo-type ratings; academic suites such as HELM score models across several dimensions; open leaderboards score open-weight models on fixed benchmarks.Which model is more capable on average?Averages and fixed tasks. Says nothing about how you use the model.AI benchmarking tools compared
AI tools and productsEditorial reviews, feature comparisons, user reviews and price lists.Which product fits a job, budget or team?Subjective and quickly out of date. Reviews rarely measure real usage.How to choose an AI tool
AI users (operators)Measured from usage telemetry. SigRank counts four token pillars from local logs and derives Υ Yield and related metrics for each operator.How efficiently does this person or workflow drive AI?Measures operating shape, not output quality, correctness, effort or intelligence.SigRank methodology

How SigRank rates operators

SigRank is the operator rating. It reads input, output, cache read and cache write from local logs and computes Υ Yield = cache_read × output / input², alongside Leverage (cache read per fresh input), Velocity (output per fresh input) and SNR (output share of fresh traffic). High Yield means context is being reused and turned into output rather than resent. See the yield cascade for the definitions.

Separately, a class ladder (IGNITER, BEARER, REFINER, SEEKER, BASE, POWER, ARCH, ARCH+) reflects accumulated experience and volume. It is not an efficiency score, so a high class and a high Yield are different facts about an operator. Explore both on the live leaderboard.

What the rating does not say: whether the work was correct, whether the code was good, how much effort it took, or how intelligent the person is. Counts only; prompts and code are never read.

FAQ

What is an AI rating?
A score that places an AI system or its user on a scale. Model ratings compare models, such as Elo-style ratings from blind human votes. Tool ratings compare products by features, price and reviews. Operator ratings, like SigRank's, compare how efficiently people use AI, measured from token counts.
How does SigRank rate AI users?
It reads four token counts (input, output, cache read, cache write) from an operator's local logs and computes Υ Yield = cache_read × output / input², plus Leverage, Velocity and SNR. Operators are placed on a leaderboard. A separate class ladder, from IGNITER up to ARCH+, reflects accumulated experience and volume, not efficiency.
Does a high SigRank rating mean the AI output is good?
No. The metrics describe how tokens flow: how much context is reused and how much output comes back per unit of fresh input. They do not measure correctness, code quality, effort, productivity or intelligence.
Can I compare two AI tools by their SigRank scores?
Not reliably. SigRank scores operators, and the same tool can produce very different cascades in different hands. Use model benchmarks to compare models, and measure your own usage to compare how you drive each tool.
Are AI ratings objective?
Only as far as the method is. Fixed benchmarks are reproducible but can be gamed or saturate; preference votes reflect voters; reviews reflect reviewers; telemetry reflects who submits it. Check what a rating measures, the sample behind it and the date before relying on it.

Related: Best AI Tools 2026 · The Tool Is the Person · How to Compare AI Operators