THE EVIDENCE BEHIND THE MODELS

Model–Benchmark.

Different tasks. Different strengths.
Explore the rankings, then go straight to the source.

— benchmark views20 entries per leaderboard, where availableWeekly research · source dates shown below

Loading benchmark rankings…

How to read these rankings

Every table keeps its benchmark version, metric and evaluation setup. Agent results measure the model and its harness together. Human-preference ratings, task completion percentages and intelligence indices are different measurements; we do not combine them into an overall score.

We publish up to 20 source-ranked entries per view, retaining source ties and system variants. When fewer entries can be verified, only those entries are shown. Source updates and our evidence checks are separate dates. If a refresh fails, the previous verified edition stays visible and its original evidence date remains unchanged.

Rankings are a starting point, not a guarantee of performance on your work. Read the original methodology, evaluation budget and limitations before choosing a model.

Explore more benchmarks on Hugging Face ↗

FOLLOW THE EVIDENCEBenchmark sourcesGo directly to the original leaderboards and their evaluation methods.11 links

Each board and ranked entry links to its specific evidence. Versions, evaluation settings, source dates and ties matter; scores from different benchmarks are not directly comparable. The linked publisher may have newer results than our saved edition.

Source links open in a new tab.

Original leaderboards11