Papers With Code Updates Support for Closed-Source Model Benchmarks

Papers With Code, operated by the Hugging Face open-source team, now provides the latest SOTA (State-of-the-Art) information through a major update.

Key updates include:

• Closed-source models support: The update now allows viewing benchmark results for models whose code is not publicly available, such as GPT-5.5 and Mythos 5.
• Customizable view settings: Users can turn closed-source model metrics on or off through settings or toggle features, allowing them to view a separate leaderboard for Open models only.
• Expanded data sources: Benchmarks are generated by parsing papers from various sources including blog posts, in addition to arXiv and Hugging Face.
• Visualization tools: Scatter plot and table visualizations are provided for each benchmark, making it easier to compare performance between models.