{"id":"arxiv:2506.21734","title":"Hierarchical Reasoning Model","date":"2025-06-26","section":"Reasoning and the \"physics\" of language models","quality":null,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/","source_url":"https://arxiv.org/abs/2506.21734","text_available":true,"authors":["Guan Wang","Jin Li","Yuhao Sun","Xing Chen"],"outgoing":[{"source":"arxiv:2506.21734","target":"arxiv:1807.03819","evidence":"id","id":"arxiv:1807.03819","title":"Universal Transformers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/1d843516d980c3404cb622a2a490edb535a3afe27a85a980335a54b5fcd2220a/"},{"source":"arxiv:2506.21734","target":"arxiv:2002.05202","evidence":"title","id":"arxiv:2002.05202","title":"GLU Variants Improve Transformer","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e052a2974593d29355ab45c301766ac6f5545ac1bc81d605e78d3adc71e9e2e4/"},{"source":"arxiv:2506.21734","target":"arxiv:2412.06769","evidence":"title","id":"arxiv:2412.06769","title":"Training Large Language Models to Reason in a Continuous Latent Space","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/68c71dc1a042dedd5e3c4d9c1ad32c9d046fd31b8dc79080bd6dff1c9800abc7/"},{"source":"arxiv:2506.21734","target":"arxiv:2501.19393","evidence":"title","id":"arxiv:2501.19393","title":"s1: Simple test-time scaling","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/2a88818f7605b71cd0160831faebd51b3274c24ffd29626b4d519390b4bffffe/"},{"source":"arxiv:2506.21734","target":"arxiv:2502.05171","evidence":"title","id":"arxiv:2502.05171","title":"Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/24a6ab865777290627cecebd3f06ea3daef01622f102d32a95f7fb2a5a42880f/"},{"source":"arxiv:2506.21734","target":"arxiv:2504.20571","evidence":"id","id":"arxiv:2504.20571","title":"Reinforcement Learning for Reasoning in Large Language Models with One Training Example","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/b2b21c3df49619588c371411dba9467a41f81b3fd4da5441898b5145bb0de120/"},{"source":"arxiv:2506.21734","target":"doi:10.1038/s41586-024-07522-w","evidence":"title","id":"doi:10.1038/s41586-024-07522-w","title":"Language is primarily a tool for communication rather than thought","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/60cfed62cd01166a972691a7c0da286223d861d23feebc5fadd3fa446168a33c/"},{"source":"arxiv:2506.21734","target":"url:pub.sakana.ai/ctm","evidence":"id","id":"url:pub.sakana.ai/ctm","title":"Continuous Thought Machines","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/2d6fdb072de0d3edb79a12bb935f300e9fe9b517f681e39459a39db71cfd44f8/"}],"incoming":[{"source":"arxiv:2509.03646","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2509.03646","title":"Emergent Hierarchical Reasoning in LLMs through Reinforcement Learning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/683657af0d699b7d854ce84d95bcd53c12e152aa738c602d0982cc0d11009f08/"},{"source":"arxiv:2510.04871","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2510.04871","title":"Less is More: Recursive Reasoning with Tiny Networks","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/4345eebe9dc7e67f82617896772f55f0970971516995c1688b6812f9ebf054f6/"},{"source":"arxiv:2512.14693","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2512.14693","title":"Universal Reasoning Model","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/fcb1a389a132d64ca93fac5551d8f8b653c5e093908cb515e2a2d901524941a7/"},{"source":"arxiv:2604.11791","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2604.11791","title":"A Mechanistic Analysis of Looped Reasoning Language Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/ae8c682d5708c6c47d65c4eb540dd5bccb877c5e332852765507128d6d1b985d/"},{"source":"arxiv:2604.21999","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2604.21999","title":"Universal Transformers Need Memory: Depth-State Trade-offs in Adaptive Recursive Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/963f6bb396c25bff462a25854c82e7d92b77803fa60c891fadd9963cc56c8fea/"},{"source":"arxiv:2605.12466","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2605.12466","title":"Solve the Loop: Attractor Models for Language and Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/eb46f81cd5718eb6b9256c583cf9dd2e97b793b4d14337e487c16285f96e2b9a/"},{"source":"arxiv:2605.19376","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2605.19376","title":"Generative Recursive Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/795bad9bb1b1a063090fbb925ba6e68bf30211de0ecffd75d56f858b28321ae7/"},{"source":"arxiv:2605.21488","target":"arxiv:2506.21734","evidence":"title","id":"arxiv:2605.21488","title":"Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/48b3beca0f016a3683df4c6c71fc4be9f13a2c12c61d4ad17d1117c6139bb11e/"},{"source":"arxiv:2606.18206","target":"arxiv:2506.21734","evidence":"title","id":"arxiv:2606.18206","title":"Fixed-Point Reasoners: Stable and Adaptive Deep Looped Transformers","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/37ea2d83a92ece4ab3ba4434af71f291d1fdc5040676f75110210f40aad3b81a/"},{"source":"arxiv:2608.09888","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2608.09888","title":"BDH-CQ: In-Context Learning with Recurrent Latent Reasoning","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/f0c77a8a0f50254e9efdf9a511e08fc2a40677bfa029864bdd0a9cbce561c666/"},{"source":"arxiv:2609.01117","target":"arxiv:2506.21734","evidence":"id","id":"arxiv:2609.01117","title":"Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/e9d4f61afc3930fa770d5ea53d3f43a5157fa4904fcdfaa514786e617905e20b/"},{"source":"arxiv:2609.33540","target":"arxiv:2506.21734","evidence":"title","id":"arxiv:2609.33540","title":"Reasoning on the Simplex: Geometric Fixed-Point Models","url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/d8882e1cc48049a4b6b9a911701665da57fcc16f9e52f2b403b1a5b334653d9c/"}],"text_parts":[{"start":0,"end":14504,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html"},{"start":14504,"end":30231,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html"},{"start":30231,"end":45585,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html"},{"start":45585,"end":51826,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html"}],"contents":[{"title":"Hierarchical Reasoning Model","offset":0,"level":1,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"Abstract","offset":32,"level":6,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"1 Introduction","offset":1587,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"2 Hierarchical Reasoning Model","offset":7443,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"Hierarchical convergence","offset":10961,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"Approximate gradient","offset":12889,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-1.html#text"},{"title":"Deep supervision","offset":18172,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"Adaptive computational time (ACT)","offset":20141,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"Inference-time scaling","offset":24295,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"Stability of Q-learning in ACT","offset":25010,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"Architectural details","offset":25896,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"3 Results","offset":28198,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"3.1 Benchmarks","offset":28509,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-2.html#text"},{"title":"3.2 Evaluation Details","offset":33038,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"3.3 Visualization of intermediate timesteps","offset":36261,"level":3,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"4 Brain Correspondence","offset":38211,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"5 Related Work","offset":43498,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"Reasoning and algorithm learning","offset":43517,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"Brain-inspired reasoning architectures","offset":45540,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-3.html#text"},{"title":"Hierarchical memory","offset":47004,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"},{"title":"6 Discussions","offset":47681,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"},{"title":"Turing-completeness of HRM","offset":47699,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"},{"title":"Reinforcement learning with chain-of-thought","offset":48646,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"},{"title":"Linear attention","offset":49497,"level":4,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"},{"title":"7 Conclusion","offset":50534,"level":2,"url":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/part-4.html#text"}],"text_code_points":51826,"source_markdown":"https://vinnik-dmitry07.github.io/favorite-papers/agents/papers/3604312101ff09ab9903e7d6a32c7b2b8a9a7ece6a9d145d04ea01c9d13ffa3a/source.md"}