Coding leaderboard
Ranked by code task pass rate, which measures whether a model can actually make code work in a real repository. The most useful board for scripts, SDK integration and refactors.
The leaderboard data has not been fetched yet. Check back later.
Where the data comes from
Scores are the pass rate on Terminal-Bench Hard, a hard measure of whether the code runs. Code style is not scored.
Scores, sample sizes, confidence intervals and prices come from the source. We do not re-score models or estimate missing numbers.
Sources: Artificial Analysis
FAQ
How is the coding board different from the overall board?
The coding board only looks at code task pass rate, while the overall board aggregates several evaluation types. A model can be strong at coding but average overall, so read both.
How often do the boards update?
The backend fetches once a day and every page load reads the latest snapshot. If a fetch fails, the previous data stays visible with its original date.
Why does a model rank so differently across boards?
The metrics differ: overall uses aggregated capability indexes, coding uses task pass rates, and image and video use blind human preference. Different metrics produce different orders.
Leaderboard data is provided by third-party public sources and is meant as a model selection reference only. It does not represent EasyAPI's own assessment of model quality. Follow the source attribution requirements when reusing it.