Reviving Papers with Code

I played a somewhat unusual role in this whole story. I was the guy who acquired in the original Papers with Code and managed them after they joined Facebook/Meta.

It was super sad to see FB/M abandon the original mission of what PwC was building towards and let the original community resource rot. During the good times, we always talked about how PwC related to HF. So, I think there is a sort poetry to PwC winding up as part of HF, where they probably always belonged. No company is perfect, but HF has been a better than average steward of open source and community resources.

For the younger folks on this thread, you probably have no real feel for just how frustratingly inefficient AI/ML research used to be before people like Robert and Ross of PwC came along to start to bring structure, sanity, and reproducibility to the information needed to work of this kind. And of course, Clem, Julien, and Thomas of HF kicked off an even bigger effort to tame the previously scattered workflow of open AI research into some sort of sane stack.

It's clear that, in 2026, what PwC could be is something much more evolved than what we were able to do back in the day. LLMs + PwC is a huge design space. I hope nielz_r and friends at HF are able to make something truly useful for the community. AI research has both gotten way easier and much harder. e.g. We have a Fable, but Anthro won't let us use it forward our science. Community resources for research are still very much needed.

Best of luck Son of PwC. May you thrive.

Hi,

Niels here from the open-source team at Hugging Face. Like many others, I was a huge fan of paperswithcode, a website which allowed to easily find the state-of-the-art (SOTA) across any domain of AI, from computer vision to language models to time-series forecasting. Sadly, that website is no longer maintained after its acquisition by Meta.

Hence, I've been working on reviving it. I obviously use AI agents to parse papers at scale and automatically generate leaderboards (for now I'm the one verifying results). So far, I've only parsed high-impact papers for which I know they're SOTA, like Qwen 3.5 and 3.6, RF-DETR for object detection, DINOv3, SOTA embedding models from the MTEB leaderboard, the Open ASR Leaderboard for automatic speech recognition models, etc.

For now, it includes the following:

> trending papers by default based on Github star velocity

> categorization by domain, e.g., [OCR](https://paperswithcode.co/tasks/ocr)

> methods, popular techniques used across AI papers, which PwC used to have as well, like [RLVR](https://paperswithcode.co/methods/rlvr) and

> eval results for high-impact papers, see e.g., Qwen 3.5 at the bottom

> leaderboards for each domain, e.g., MMTEB or COCO val 2017

> conferences, like [CVPR 2026](https://paperswithcode.co/conferences/cvpr-2026)

> support for citation counts (you can also see the most cited papers by domain!)

> automated linked Github, project page URLs, and artifacts (+ multiple repos are supported on a paper page)

> support for external papers beyond Arxiv, see e.g., [DeepSeek v4](https://paperswithcode.co/paper/82956)

> Harness reports for coding agent benchmarks, e.g., Terminal Bench

> "Sign in with HF" and Storage Buckets are used to store humbnails, paper PDFs, and overall data backups.

I'm curious about your feedback + feature requests!

Try it at https://paperswithcode.co

Shame about the name, it feels better suited to a more general curated repo/hall-of-fame of papers in any field that come with easily rerunnable code to reproduce the paper’s results, or try out different datasets, or similar.

Is no one tired else of these repetitive, obviously Claude-made webdesigns?

This is a bit off-topic (though tangentially related) - does anyone remember a similar blog where the author would do something like a "5 minute paper" review, i.e. they'd discuss findings and try to communicate the main point? It was usually a paper per week, mostly CompSci / maths papers IIRC

The fact that this even needs to exist means that our academic conferences aren't prioritizing the right things. Papers without code should be desk rejected.

Yes please! I have been frustrated with the state of object detection models especially. Everyone claims SOTA. So you end up having to test manually to find out which one actually is. And unlike LLM's, it should be pretty easily quantifiable.

Is there an RSS feed?

Bring it back! Sing it back!