Papers with Code is Back! The Revival Every ML Engineer and Researcher Has Been Waiting For
If you’ve ever spent hours hunting for the latest SOTA model in speech recognition, text retrieval, or any other AI subfield, you know the pain. The original Papers with Code was that magical one-stop shop: a clean, addictive leaderboard of papers, code, datasets, and benchmarks that let you instantly see who was winning where. Then it quietly faded after Meta’s 2019 acquisition. For years, the community missed it.
Well, it’s officially resurrected — and better than ever.
Welcome to paperswithcode.co, the brand-new revival spearheaded by Hugging Face’s Niels Rogge in close collaboration with Meta AI and the original creators (Robert Stojnic and Ross Taylor). Built from the ground up with heavy use of AI agents that parse papers at scale (with human verification for quality), the new site brings back everything we loved and adds the modern features the fast-moving AI field actually needs in 2026.
Why It Feels Like Coming Home
Remember scrolling through trending papers and instantly spotting which library or technique just crushed a benchmark? That magic is back. The site now surfaces the freshest research automatically, so you’re never stuck refreshing arXiv manually. And yes — the classic benchmark tables are fully restored, complete with clean leaderboards across dozens of industries and tasks.
Take the Massive Text Embedding Benchmark (MMTEB) as a perfect example. The current leaderboard (https://paperswithcode.co/benchmark/mmteb) shows a massive Microsoft model sitting at the very top, but among the more widely used vendors, Google’s embeddings, Anthropic’s Claude family, and OpenAI’s latest offerings are all trading blows in the upper ranks. Exactly the kind of at-a-glance comparison the community has been craving.
New Superpowers for the AI Era
The revival didn’t just copy the old site — it leveled it up:
- Trending by Research Direction
Clear visual trends show exactly where the big labs are pouring their energy right now. You can watch entire fields heat up in real time.
- Niche Feeds You Can Actually Subscribe To
World Models, Agents, Reinforcement Learning, Robotics, and dozens of other specialized topics now have dedicated, high-signal feeds. Subscribe once and never miss the next breakthrough in your corner of AI.
- The brand-new “Methods” Tab
This one is pure gold for engineers and researchers. Want to see every paper that uses RLVR? Speculative decoding? MoE routing? Just click the technique and get the full list of implementations, results, and code. It’s the kind of deep, method-focused navigation that used to require custom scripts or Twitter threads.
The design? It has that unmistakable “vibe-coding” hacker aesthetic — functional, no-nonsense, with a bit of retro flair. Some might call it quirky, but it feels right at home for the people who actually ship models.
Also read:
- Pizza Hut’s $100 Million AI Meltdown: How “Smart” Delivery Tech Delivered Cold Pizzas and a Lawsuit
- Meanwhile, OpenAI Just Killed Another Hundred Startups
- Andon Labs Let Four AIs Run Real 24/7 Internet Radio Stations for Six Months. The Result Was Pure, Unfiltered Mayhem.
- Researchers Stirred Up Chaos for AI Agents—And Watched Them Lose Their Minds in Record Time. Spoiler: It Was Brutal.
Who This Is For
If you run an ML team, this is now your default morning dashboard.
If you’re picking a new embedding model, agent framework, or robotics stack for a product, the benchmarks and code links are right there.
If you’re a researcher or indie hacker, the niche subscriptions and Methods tab mean you’ll actually stay on top of your field without drowning in noise.
In short: Papers with Code is alive again, fresher, more focused, and more useful than ever. The age of research is back — and the community finally has its favorite map again.
Go check it out: paperswithcode
Bookmark it. Subscribe to your niches. Watch the leaderboards move.
The AI research game just got a whole lot more fun.