{"slug": "classical-energies-learn-to-reason-1kx-faster-than-looped-models", "title": "Classical Energies Learn to Reason 1k× Faster Than Looped Models", "summary": "A paper submitted to arXiv on 12 Nov 2026 reports that CRAM, a learned version of Hopfield and Tank's 1980s energy model, solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles after 42 ms and 132 ms of training on one GPU, at least 1,000x faster than looped reasoning models on those tasks. CRAM also matches baseline looped models at their highest Sudoku-Extreme accuracy using at least 282x fewer inference FLOPs. The authors conclude that energies are a promising direction for training reasoning from data and caution against using trivially solvable benchmarks like Sudoku and Maze to test the scalability of looped reasoning techniques.", "body_md": "# Computer Science > Artificial Intelligence\n\n  [Submitted on 12 Nov 2026]\n\n# Title:Classical Energies Learn to Reason 1,000x Faster than Looped Models\n\n[View PDF](https://arxiv.org/pdf/2611.17309)\n\n[HTML (experimental)](https://arxiv.org/html/2611.17309v1)\n\nAbstract:Looped models have emerged as a prominent paradigm for reasoning, with successes on simple reasoning puzzles such as Sudoku-Extreme and Maze-Unique. However, these models can be unstable, slow to train, and use many parameters and FLOPs to solve simple puzzles. Descent on an energy not only mitigates these issues but also offers advantages: built-in early stopping, compute that adapts to problem difficulty, fast training without differentiating through a solver, and interpretable weights that learn the constraints of the puzzle. Here we demonstrate this with a learned version of Hopfield & Tank's energy from the 80s, which we term CRAM, solves 99.996% of Sudoku-Extreme and 100% of Maze-Unique test puzzles after 42 ms and 132 ms of training on one GPU, at least 1,000x faster than looped models on these tasks. CRAM also matches baseline looped models at their highest accuracy on Sudoku-Extreme using at least 282x fewer inference FLOPs. Our work suggests two things. First, that energies are an extremely promising direction for training reasoning from data. Second, a caution against using trivially solvable benchmarks like Sudoku and Maze to test the scalability of looped reasoning techniques.\n    \n\n### Additional Features\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer \n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers \n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps \n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations \n\n*(*[What are Smart Citations?](https://www.scite.ai/))\n# Code, Data and Media Associated with this Article\n\nalphaXiv \n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers \n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub \n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub \n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face \n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast \n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))\n# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower \n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender \n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))\n# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [**Learn more about arXivLabs**](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/classical-energies-learn-to-reason-1kx-faster-than-looped-models", "canonical_source": "https://arxiv.org/abs/2611.17309", "published_at": "2026-11-18 05:12:47+00:00", "updated_at": "2026-11-18 05:45:50.742569+00:00", "lang": "en", "topics": ["artificial-intelligence", "machine-learning", "ai-research", "large-language-models"], "entities": ["arXiv", "CRAM", "Hopfield and Tank", "Sudoku-Extreme", "Maze-Unique"], "also_reported_by": [], "takeaway": "CRAM, a learned Hopfield-Tank energy model, reaches 99.996% on Sudoku-Extreme and 100% on Maze-Unique after 42 ms and 132 ms of single-GPU training, at least 1,000x faster than looped models, while matching their top Sudoku accuracy with at least 282x fewer inference FLOPs.", "recommendation": {"advice": "Don't judge looped reasoning techniques on Sudoku or Maze alone; the paper shows an energy model solves them at 1,000x lower training cost, so test scalability on harder benchmarks.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/classical-energies-learn-to-reason-1kx-faster-than-looped-models", "markdown": "https://wpnews.pro/news/classical-energies-learn-to-reason-1kx-faster-than-looped-models.md", "text": "https://wpnews.pro/news/classical-energies-learn-to-reason-1kx-faster-than-looped-models.txt", "jsonld": "https://wpnews.pro/news/classical-energies-learn-to-reason-1kx-faster-than-looped-models.jsonld"}}