{"paper":{"title":"What Matters for Latent Reasoning with Flow Matching","arxiv_id":"2610.06666","primary_cat":"cs.LG","submitted_at":"2026-10-05T16:42:40+00:00","version_compiled":"v1"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v1","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Latent reasoning lets a large language model (LLM) think in a continuous space and verbalize only the answer.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.06,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.48,"id":"s1","own":false,"role":"contribution","text":"We argue that an effective latent thought must meet five requirements: it should be useful, helping produce the correct answer rather than merely changing it, diverse, so that resampling yields different reasoning trajectories, explainable, so that a decoded chain of thought (CoT) reflects reasoning the answer actually follows, refinable with more inference compute, and efficient, costing less than an explicit CoT at comparable accuracy.","hedge":{"p":0.91,"label":"asserted"},"own_p":0.27,"main_claim_p":0.15,"quantitative":true,"quantitative_p":0.68,"role_probabilities":{"gap":0.2,"method":0.02,"result":0.0,"background":0.29,"limitation":0.0,"future_work":0.0,"contribution":0.48,"interpretation":0.01}},{"i":2,"p":0.88,"id":"s2","own":false,"role":"gap","text":"Current methods rarely meet these requirements: they learn shortcuts from the question, distill the explicit CoT into their weights, or imitate it one token at a time.","hedge":{"p":0.94,"label":"asserted"},"own_p":0.09,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.12,"role_probabilities":{"gap":0.88,"method":0.0,"result":0.0,"background":0.12,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":3,"p":0.66,"id":"s3","own":true,"role":"contribution","text":"We focus on flow matching in a learned latent space, the family we argue is best placed to meet them, and identify the training choices that make it work.","hedge":{"p":0.78,"label":"asserted"},"own_p":0.51,"main_claim_p":0.12,"quantitative":false,"quantitative_p":0.32,"role_probabilities":{"gap":0.0,"method":0.33,"result":0.0,"background":0.01,"limitation":0.0,"future_work":0.0,"contribution":0.66,"interpretation":0.0}},{"i":4,"p":0.83,"id":"s4","own":true,"role":"contribution","text":"The result is Flow-based Latent Reasoning (FLaRe), a simple recipe covering what the latent space encodes and how to shape it, where to train the flow, how to read out the answer, and a final stage of training on the model's own verified thoughts.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.91,"main_claim_p":0.28,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.17,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.83,"interpretation":0.0}},{"i":5,"p":0.98,"id":"s5","own":true,"role":"result","text":"A probe for each requirement shows that FLaRe improves on prior latent methods in all five.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.92,"main_claim_p":0.26,"quantitative":true,"quantitative_p":0.61,"role_probabilities":{"gap":0.0,"method":0.01,"result":0.98,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.01,"interpretation":0.0}},{"i":6,"p":1.0,"id":"s6","own":true,"role":"result","text":"It also compares favorably with them on arithmetic benchmarks, while reaching 97% of the accuracy of explicit CoT at a quarter of its latency.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.35,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}}]},"compiled":{"main_claim":{"p":0.71,"text":"A probe for each requirement shows that FLaRe improves on prior latent methods in all five.","location":"abstract s5","hedge_strength":{"p":1.0,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":5744,"cost_usd_reported":0.0},"compiled_at":"2026-10-06","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.06,"states_limitation":0.22,"human_participants":0.03,"controlled_comparison":0.81,"randomized_assignment":0.01,"states_quantitative_result":0.99,"compares_with_prior_results":0.97,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":0.85,"value":"observation or experiment","probabilities":{"argument":0.03,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":0.85,"simulation_or_computation":0.12}},"contribution_type":{"p":1.0,"value":"method or tool","probabilities":{"review":0.0,"method_or_tool":1.0,"empirical_study":0.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":0.99,"text":"The result is Flow-based Latent Reasoning (FLaRe), a simple recipe covering what the latent space encodes and how to shape it, where to train the flow, how to read out the answer, and a final stage of training on the model's own verified thoughts.","location":"abstract s4"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v1"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":7,"sentences_in_abstract":7},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448,"review_comments_judged":{"judged_by":["gpt-5.6-luna","grok-4.7"],"judged_against":"Pith's own text copy of each paper: text extracted from the PDF, the first 90,000 characters","measured_on":"2026-10-04","accurate":113,"decided":149,"sampled_reviews":200,"split_by_lane":false,"by_lane":"https://pith.science/accuracy#by-lane"}}}