mlsanityData to 7 Oct 2026

Quantizing Recursive Reasoning Models

arXiv 2607.16237 ↗First posted 25 Jun 2026cs.LG cs.AIcited by 2

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

Recursive reasoning models solve hard puzzles by applying compact, weight-tied blocks over many refinement steps. Because these blocks are reused many times, quantizing them creates a unique dynamical problem: the quantization error is incurred at every step. While 8-bit quantization (integer or float) preserves accuracy, moving to a per-tensor 4-bit format causes a systematic bias to accumulate. The ensuing drift catastrophically degrades exact-solution accuracy on Sudoku from 84.1% to 0.0% (only ~25% of cells correct). In this work, we show that this collapse is caused by activation-scaling granularity rather than bit-width or number format. Crucially, moving to per-block scaling completely restores the transition. To implement this, we apply MXInt4, a blockwise integer activation format, to recursive reasoning models. It is competitive with blockwise float formats on our tasks, while keeping integer elements and power-of-two block scales. Finally, recursion depth and reuse modulate quantization sensitivity, with the deepest architecture we test (the EqR equilibrium model) the most sensitive. Yet blockwise scaling overcomes this vulnerability, staying robust across these architectures and transferring to the open-ended ARC-AGI benchmark.

Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 51 of its 84 references that are in the map. First posted on arXiv 25 Jun 2026.

Builds on

51 of 84 references.

Closest ancestors · 13

Same direction, newest first

  1. 2026Equilibrium Reasoners: Learning Attractors Enables Scalable Reasoning20 May
  2. 2026LAQuant: A Simple Overhead-free Large Reasoning Model Quantization by Layer-wise Lookahead Loss9 May
  3. 2026Recursive Models for Long-Horizon Reasoning2 Mar
  4. 2026Tiny Recursive Reasoning with Mamba-2 Attention Hybrid12 Feb
  5. 2026QuantLRM: Quantization of Large Reasoning Models via Fine-Tuning Signals31 Jan
  6. 2026What Makes Low-Bit Quantization-Aware Training Work for Reasoning LLMs? A Systematic Study21 Jan
  7. 2025Accelerating Training Speed of Tiny Recursive Models with Curriculum Guided Adaptive Recursion11 Nov
  8. 2025Less is More: Recursive Reasoning with Tiny Networks6 Oct
  9. 2025Hierarchical Reasoning Model26 Jun
  10. 2025Quantization Meets Reasoning: Exploring and Mitigating Degradation of Low-Bit LLMs in Mathematical Reasoning16 May
  11. 2025Quantization Hurts Reasoning? An Empirical Study on Quantized Reasoning Models7 Apr
  12. 2025When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models2 Apr
1 more
  1. 2025Quantization Meets Reasoning: Exploring LLM Low-Bit Quantization Degradation for Mathematical Reasoning6 Jan

Other references · 38

  1. 2013Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computationcited by 2,031
  2. 2024QuaRot: Outlier-Free 4-Bit Inference in Rotated LLMscited by 598
  3. 2016Adaptive Computation Time for Recurrent Neural Networkscited by 479
  4. 2018PACT: Parameterized Clipping Activation for Quantized Neural Networkscited by 456
  5. 2021A White Paper on Neural Network Quantizationcited by 453
  6. 2024SpinQuant: LLM quantization with learned rotationscited by 446
  7. 2023Microscaling Data Formats for Deep Learningcited by 132
  8. 2025Pretraining Large Language Models with NVFP4cited by 67
  9. 2025Quartet: Native FP4 Training Can Be Optimal for Large Language Modelscited by 37
  10. 2025FP4 All the Way: Fully Quantized Training of LLMscited by 37
  11. 2025Bridging the Gap Between Promise and Performance for Microscaling FP4 Quantizationcited by 30
  12. 2025Oscillation-Reduced MXFP4 Training for Vision Transformerscited by 20
26 more
  1. 2023Integer or Floating Point? New Outlooks for Low-Bit Quantization on Large Language Modelscited by 18
  2. 2025Block Rotation is All You Need for MXFP4 Quantizationcited by 18
  3. 2024AMXFP4: Taming Activation Outliers with Asymmetric Microscaling Floating-Point for 4-bit LLM Inferencecited by 17
  4. 2025INT v.s. FP: A Comprehensive Study of Fine-Grained Low-bit Quantization Formatscited by 16
  5. 2026Generative Recursive Reasoningcited by 16
  6. 2026Probabilistic Tiny Recursive Modelcited by 16
  7. 2016Effective Quantization Methods for Recurrent Neural Networkscited by 14
  8. 2026ARCQuant: Boosting NVFP4 Quantization with Augmented Residual Channels for LLMscited by 14
  9. 2016Recurrent Neural Networks With Limited Numerical Precisioncited by 12
  10. 2025Benchmarking Post-Training Quantization in LLMs: Comprehensive Taxonomy, Unified Evaluation, and Comparative Analysiscited by 12
  11. 2018Effective Quantization Approaches for Recurrent Neural Networkscited by 10
  12. 2026Benchmarking Post-Training Quantization of Large Language Models under Microscaling Floating Point Formatscited by 8
  13. 20214-bit Quantization of LSTM-based Speech Recognition Modelscited by 7
  14. 2026Is Finer Better? The Limits of Microscaling Formats in Large Language Modelscited by 7
  15. 2026HRM-Text: Efficient Pretraining Beyond Scalingcited by 7
  16. 2022An Empirical Study of Low Precision Quantization for TinyMLcited by 6
  17. 2026Tiny Autoregressive Recursive Modelscited by 6
  18. 2026Unveiling the Potential of Quantization with MXFP4: Strategies for Quantization Error Reductioncited by 5
  19. 2026BATQuant: Outlier-resilient MXFP4 Quantization via Learnable Block-wise Optimizationcited by 5
  20. 2018Precision Highway for Ultra Low-Precision Quantizationcited by 3
  21. 2023Quantized Neural Networks for Low-Precision Accumulation with Guaranteed Overflow Avoidancecited by 3
  22. 2025Q-Sched: Pushing the Boundaries of Few-Step Diffusion Models with Quantization-Aware Schedulingcited by 3
  23. 2026Quantization Robustness of Monotone Operator Equilibrium Networkscited by 3
  24. 2026Q-Drift: Quantization-Aware Drift Correction for Diffusion Model Samplingcited by 3
  25. 2025Quant-Trim in Practice: Improved Cross-Platform Low-Bit Deployment on Edge NPUscited by 2
  26. 2025BitSkip: An Empirical Analysis of Quantization and Early Exit Composition in Transformerscited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026Q-DEQ: Discrete Solving and Quantization for Deep Equilibrium Models in Time Series Forecasting under Edge Deployment Coding Constraintscited by 0
  2. 2026A Tilted Bowl Is Not a Slippery Slope: Compressing Looped Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Recursive Vision Language Models for General Symbolic Reasoning
  2. 2026Energy-guided Recursive Model
  3. 2026What Survives When You Compress a Recursive Reasoner for the Edge?
  4. 2026Search, Fail, Recover: A Training Framework for Correction-Aware Reasoning
  5. 2026What Limits Recursive Reasoning Models: Optimization, Architecture and Test-Time Scaling
  6. 2026Recurrent Reasoning on Symbolic Puzzles with Sequence Models
  7. 2025Liquid Reasoning Transformers: A Sudoku-Based Prototype for Chess-Scale Algorithmic Tasks
  8. 2026Reasoning on the Simplex: Geometric Fixed-Point Models