Introducing Papers Without Code [P]
![Introducing Papers Without Code [P]](/_next/image?url=https%3A%2F%2Fpreview.redd.it%2F9rz2r3ffcf6h1.png%3Fwidth%3D140%26height%3D73%26auto%3Dwebp%26s%3D5b975131682d6d537cdad294296d3173ba6e334a&w=3840&q=75)
![Introducing Papers Without Code [P]](https://preview.redd.it/9rz2r3ffcf6h1.png?width=140&height=73&auto=webp&s=5b975131682d6d537cdad294296d3173ba6e334a "Introducing Papers Without Code [P]") | Hi, Niels here from the open-source team at Hugging Face. I've recently relaunched paperswithcode.co as a source for finding the state of the art (SOTA) across various AI domains, from 3D generation to AI agents. This is done by automatically parsing research papers published on arXiv/Hugging Face, enabling leaderboards to be created. See BrowseComp below as an example (a scatter plot and a table are available for each benchmark). - Scatter plot (you can hover over the dots to see the models): - Table: As you can see, I've added support for viewing evals for closed-source models, too, given that many benchmarks are nowadays dominated by them, like GPT-5.5 and Mythos 5. You can always disable viewing closed-source evals with a toggle or in your PwC settings: When you turn them off, here's what the open model leaderboard looks like: Closed-source papers are treated as regular "papers", although they can be any source, like a blog post (given that PwC supports submitting any source beyond arXiv). See the GPT-5.5 or Mythos 5 papers as examples, with their evals at the bottom. Notice the "closed" tag on their evals. Hence, you could jokingly call these "papers without code". Let me know what you think of this, and whether anything needs to be changed or added! Kind regards, [link] [comments] |
Want to read more?
Check out the full article on the original site