Data
Sources and licences.
Every figure on the site names the page or dataset it came from. The full licence text lives here, not under every table.
Data: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from https://epoch.ai/benchmarks. Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Scores are shown as percentages, rounded to one decimal.
Human-preference ratings: LMArena Text Arena leaderboard (lmarena.ai), via the lmarena-ai/leaderboard-dataset on Hugging Face, licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Ratings rounded to whole numbers.
LiveBench 2.0 results from https://livebench.ai (White et al., LiveBench). Benchmark code licensed Apache-2.0. A LiveBench number is shown only when Epoch publishes it for that model.
List prices come from the seller’s page or from models.dev (MIT), including context, modalities and knowledge cutoff when that catalog publishes them. LiteLLM’s price file is MIT and does not publish a weights licence or a cutoff, so those lines are omitted. Settled list-price history comes from ModelPriceWatch (CC-BY-4.0). A missing day stays missing.
Hugging Face download, like and trending counts are from the public Hub API (huggingface.co/models). GitHub star counts and release tags are from the public REST API for a fixed list of open-source tools. Both sections stay hidden when the request returns nothing.
Vendor-reported scores are taken from each developer’s published model card or launch post. They were not independently verified by PowerfulBot, and they are not mixed into an independent ranking.
Maker logos are self-hosted from lobe-icons (MIT) with a Simple Icons (CC0) fallback. They identify the maker. See the trademark note.
Illustration: PowerfulBot. News covers are original artwork made for this site. When an allowlisted publisher page includes an og:image, that image is shown instead and credited on the card.
How a score gets onto a board is in the methodology.