Knowledge Base
Direct answers on Superrank methodology, Bayesian formulas, community ratings, and platform policies.
Superrank ranks AI models, agents, benchmarks, companies and people on verifiable evidence, with a Superrank Rating, user ratings and popularity. The platform delivers mathematically transparent, reproducible AI evaluations without sponsored placements, pay-to-play rankings, or synthetic marketing reviews, providing practitioners with an objective source of truth for the entire artificial intelligence landscape.
Rankings are calculated deterministically using Bayesian shrinkage: W = (v / (v + m)) * R + (m / (v + m)) * C, where v is the vote count, R is the entity's average user rating (1.0 to 10.0 scale), m = 25 is the minimum threshold weight, and C = 7.0 is the catalog prior mean. Composite rankings combine verified benchmark performance (40%), Bayesian community ratings (40%), and pricing & value index (20%).
The Superrank Rating is a Bayesian-adjusted quality score on an IMDb-style 1 to 10 scale. To prevent entities with only a few positive votes from unfairly topping the leaderboard, an entity must accumulate at least 25 verified community ratings before its Superrank Rating is computed and published on public ranking charts. If an entity has fewer than 25 ratings, we honestly display 'Not enough ratings yet'.
All benchmark evaluations are attributed with an explicit provenance badge. Self-reported benchmarks from vendor whitepapers are marked 'Self-Reported'. Evaluations reproduced by independent academic labs or third-party audit platforms (such as LMSYS Chatbot Arena or SWE-bench) are marked 'Independently Audited'.
Signed-in members can rate any AI model, autonomous agent, company, benchmark, or researcher on an IMDb-style 1 to 10 scale by clicking the 'Rate' button on the entity's page. Community members can update or delete their rating at any time, which recalculates the Bayesian rating aggregate and updates leaderboard standings.
Popularity is a normalized percentile metric measuring 30-day user engagement across Superrank. It tracks verified pageviews, rating activity, watchlist additions, and comparison frequency across the catalog. Popularity measures community velocity and interest but remains completely separate from the Bayesian Superrank Rating, ensuring popularity never skews objective quality evaluations.
Ratings require an authenticated member account. We enforce sliding-window IP rate limits, automated sybil detection, and conflict-of-interest disclosure flags. Reviews are audited by automated moderation filters and community flag reports.
Click the 'Add to Watchlist' button with the bookmark icon on any entity or ranking card. You can view, filter, and organize all your saved entities under /watchlist.
Yes. Navigate to /lists to build public or private custom lists (such as 'Top Coding LLMs for 2026' or 'Best Local Offline Models'). You can share list links or embed them on external websites.
Community members and developers can submit foundation models or autonomous agents through the /submit portal. Submissions require verifiable primary source documentation, official model weights or API references, license classification, and reproducible benchmark evaluations. Every entry undergoes rigorous editorial review to guarantee data provenance before publication to the catalog.
Creators, research lab representatives, and engineers can navigate to their profile page and click 'Claim Profile', or visit /claim/organization/new or /claim/person/new. Proof of affiliation (such as an official domain email or GitHub commit history) is verified within 48 hours.
Superrank datasets are made freely available under Open Data Commons Open Database License (ODbL) and CC-BY 4.0. You are free to share, analyze, and build upon our datasets provided you cite Superrank.
Review our deterministic mathematical specifications or reach our research team.
