Latent Recurrent Thoughts: Recurrent Refinement of Proposed Latents for Reasoning with Frozen LLMs
On the map: LLM reasoning → Test-time compute and efficient thinking
Abstract (arXiv)
Chain-of-thought reasoning unfolds in discrete token space: each step is committed as text, errors propagate, and eliciting good traces presupposes traces to imitate. Reasoning instead in a model's continuous representation space - where intermediate states are vectors rather than words - sidesteps these constraints, but leaves open how those latent states should be computed. We approach this along two axes. First, we keep a large language model (LLM) frozen and use it for what it is already good at - modeling and decoding sequences - while a small auxiliary network supplies continuous latent thoughts as input. Second, we produce those latents by recurrence: a tiny recurrent reasoner refines them over many steps, decoupling the depth of computation from the size of the model, so that the latents are a product of iterative processing rather than a single forward pass. We instantiate this as Latent Recurrent Thoughts (LRT): a task-dedicated proposer supplies base latents, a recurrent reasoner refines them through bounded residual corrections, and the frozen LLM decodes the answer. On symbolic reasoning with answer supervision but no reasoning traces (Countdown-4, Sudoku) and on natural-language reasoning (HumanEval, MBPP, StrategyQA), LRT substantially outperforms prior frozen-decoder continuous-space reasoning methods under an identical decoder, prompt, data, and training budget, and outperforms non-thinking-mode chain-of-thought prompting on the same backbone at a small fraction of its inference compute.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 40 of its 49 references that are in the map. First posted on arXiv 1 Sep 2026.
Builds on
40 of 49 references.
Closest ancestors · 13
Same direction, newest first
- 2026Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning20 May
- 2025Less is More: Recursive Reasoning with Tiny Networks6 Oct
- 2025A Survey on Latent Reasoning8 Jul
- 2025Hierarchical Reasoning Model26 Jun
- 2025Think Silently, Think Fast: Dynamic Latent Compression of LLM Reasoning Chains22 May
- 2025SoftCoT++: Test-Time Scaling with Soft Chain-of-Thought Reasoning16 May
- 2025Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach7 Feb
- 2024Compressed Chain of Thought: Efficient Reasoning Through Dense Representations17 Dec
- 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters6 Aug
- 2024Stream of Search (SoS): Learning to Search in Language1 Apr
- 2023Think before you speak: Training Language Models With Pause Tokens3 Oct
- 2023Tree of Thoughts: Deliberate Problem Solving with Large Language Models17 May
1 more
Other references · 27
- 2020Language Models are Few-Shot Learnerscited by 32,199
- 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,084
- 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
- 2025Qwen3 Technical Reportcited by 8,195
- 2021Evaluating Large Language Models Trained on Codecited by 8,053
- 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,601
- 2022Large Language Models are Zero-Shot Reasonerscited by 4,323
- 2019Parameter-Efficient Transfer Learning for NLPcited by 4,195
- 2021Program Synthesis with Large Language Modelscited by 3,509
- 2023Graph of Thoughts: Solving Elaborate Problems with Large Language Modelscited by 873
- 2021Show Your Work: Scratchpads for Intermediate Computation with Language Modelscited by 871
- 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
15 more
- 2018Universal Transformerscited by 613
- 2019Deep Equilibrium Modelscited by 603
- 2024From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Stepcited by 225
- 2023Looped Transformers as Programmable Computerscited by 224
- 2024Let's Think Dot by Dot: Hidden Computation in Transformer Language Modelscited by 186
- 2023Implicit Chain of Thought Reasoning via Knowledge Distillationcited by 165
- 2024Beyond Autoregression: Discrete Diffusion for Complex Reasoning and Planningcited by 135
- 2025Soft Thinking: Unlocking the Reasoning Potential of LLMs in Continuous Concept Spacecited by 109
- 2025Reasoning by Superposition: A Theoretical Perspective on Chain of Continuous Thoughtcited by 73
- 2025Reasoning Beyond Language: A Comprehensive Survey on Latent Chain-of-Thought Reasoningcited by 72
- 2024Learning Iterative Reasoning through Energy Diffusioncited by 36
- 2022Learning Iterative Reasoning through Energy Minimizationcited by 32
- 2020Compositional Visual Generation and Inference with Energy Based Modelscited by 25
- 2026Generative Recursive Reasoningcited by 16
- 2025Think Consistently, Reason Efficiently: Energy-Based Calibration for Implicit Chain-of-Thoughtcited by 3
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026Think Deep, Speak Once: Relit, A Recursive Latent Implicit Transformer Framework
- 2025Do Latent Tokens Think? A Causal and Adversarial Analysis of Chain-of-Continuous-Thought
- 2026A Mechanistic Analysis of Looped Reasoning Language Models
- 2026Tyler: Typed Latent Reasoning for Language Models -- When to Think, What to Compute, and How Much to Allocate
- 2025Encode, Think, Decode: Scaling test-time reasoning with recursive latent thoughts
- 2026Thinking in Latents: Adaptive Anchor Refinement for Implicit Reasoning in LLMs
- 2025Beyond Chains of Thought: Benchmarking Latent-Space Reasoning Abilities in Large Language Models
- 2026What Matters for Latent Reasoning with Flow Matching