Electricity Bench grades new AI models on private, niche task suites and publishes the results here: scorecards, an AI-tools directory, and news.
Most public leaderboards aggregate the same handful of benchmarks that every lab already optimizes for. We do something narrower and, we think, more useful. We keep our own task suites private, run them against each newly released model, and publish only the graded outcome. The tasks stay behind the curtain so they cannot be trained against; the numbers come out in the open.
What you can expect
When a notable model ships, we run it through the suites and publish a scorecard: a per-suite grade with capability, cost, and latency, plus any provider-claimed or third-party numbers kept clearly labeled as context and never blended into our own grade.
What is live today
The directory, methodology, and this news section are live. The first graded scorecards are in preparation and will appear here as the suites finish their calibration runs. Subscribe below and you will get each new model's scorecard the day it drops.