Hugging Face publishes CWE-bench v2 with a public leaderboard
Hugging Face has published CWE-bench v2, a new version of its CWE-bench benchmark, under the title "CWE-bench v2: More coverage, quality, and transparency." A leaderboard for the v2 public set is available as a Hugging Face Space.
- Hugging Face published "CWE-bench v2: More coverage, quality, and transparency" on its blog.
- The post is categorised as a Benchmark and links to one dataset.
- A Space named CWE-bench Leaderboard hosts results for the CWE-bench v2 public set.
- The available source text gives no details on task counts, methodology or model scores.
What Hugging Face has released
The post appeared on the Hugging Face blog on 3 November 2026 and is filed under the Benchmark category. Its title names three goals for the second version: more coverage, better quality and greater transparency.
The page links to one dataset and one Space. The Space is called CWE-bench Leaderboard and is described as a "Leaderboard for the CWE-bench v2 public set," according to the Hugging Face page.
What the source does not say
The text available for this article contains only the headline, the category and the references to the dataset and the leaderboard. It does not describe how the benchmark was built, how many tasks it contains, which models were tested or how they scored. It also does not explain what changed between the first and second versions. This article therefore makes no claims about those points.
The description of the leaderboard refers to a "public set", which indicates that the results shown there relate to the publicly available portion of CWE-bench v2.
What it means
Based on the title, Hugging Face is positioning the update as an improvement in the breadth of the benchmark, the quality of its content and the openness of how it is run. The public leaderboard gives developers and researchers a place to see how submissions compare on the v2 public set.
Details on methodology and results will be in the full blog post and the linked dataset, which readers should consult directly for specifics.
Hugging Face: CWE-bench v2: More coverage, quality, and transparency
