mlsanityData to 14 Oct 2026

Locally Confident, Globally Stuck: The Quality-Exploration Dilemma in Diffusion Language Models

arXiv 2604.00375 ↗First posted 1 Apr 2026cs.CLcited by 5

On the map: LLM reasoning → Diffusion reasoning and test-time scaling

also in 1 more
Abstract (arXiv)

Diffusion large language models (dLLMs) theoretically permit token decoding in arbitrary order, a flexibility that could enable richer exploration of reasoning paths than autoregressive (AR) LLMs. In practice, however, random-order decoding often hurts generation quality. To mitigate this, low-confidence remasking improves single-sample quality (e.g., Pass@1) by prioritizing confident tokens, but it also suppresses exploration and limits multi-sample gains (e.g., Pass@k), creating a fundamental quality--exploration dilemma. In this paper, we provide a unified explanation of this dilemma. We show that low-confidence remasking improves a myopic proxy for quality while provably constraining the entropy of the induced sequence distribution. To overcome this limitation, we characterize the optimal distribution that explicitly balances quality and exploration, and develop a simple Independent Metropolis--Hastings sampler that approximately targets this distribution during decoding. Experiments across a range of reasoning benchmarks including MATH500, AIME24/25, HumanEval, and MBPP show that our approach yields better exploration-quality tradeoff than both random and low-confidence remasking.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Placed in Diffusion reasoning and test-time scaling, one of 17 directions in LLM reasoning. Builds on 42 of its 53 references that are in the map. First posted on arXiv 1 Apr 2026.

Builds on

42 of 53 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025d-TreeRPO: Towards More Reliable Policy Optimization for Diffusion Language Models10 Dec
  2. 2025Beyond Surface Reasoning: Unveiling the True Long Chain-of-Thought Capacity of Diffusion Large Language Models10 Oct
  3. 2025d1: Scaling Reasoning in Diffusion Large Language Models via Reinforcement Learning16 Apr

Other references · 39

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  3. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  4. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  5. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  6. 2021Program Synthesis with Large Language Modelscited by 3,637
  7. 2023Is Your Code Generated by ChatGPT Really Correct? Rigorous Evaluation of Large Language Models for Code Generationcited by 1,391
  8. 2025Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?cited by 982
  9. 2025Large Language Diffusion Modelscited by 912
  10. 2024Simple and Effective Masked Diffusion Language Modelscited by 848
  11. 2023Discrete Diffusion Modeling by Estimating the Ratios of the Data Distributioncited by 745
  12. 2024Simplified and Generalized Masked Diffusion for Discrete Datacited by 626
27 more
  1. 2025Dream 7B: Diffusion Large Language Modelscited by 544
  2. 2022A Continuous Time Framework for Discrete Denoising Modelscited by 451
  3. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 420
  4. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Modelscited by 409
  5. 2024Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Datacited by 381
  6. 2025Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusionscited by 214
  7. 2024Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Samplingcited by 208
  8. 2025Seed Diffusion: A Large-Scale Diffusion Language Model with High-Speed Inferencecited by 178
  9. 2024The pitfalls of next-token predictioncited by 160
  10. 2025The Diffusion Dualitycited by 135
  11. 2025Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmaskingcited by 131
  12. 2025Reasoning with Sampling: Your Base Model is Smarter Than You Thinkcited by 89
  13. 2025KLASS: KL-Guided Fast Inference in Masked Diffusion Modelscited by 47
  14. 2025Wide-In, Narrow-Out: Revokable Decoding for Efficient and Effective DLLMscited by 46
  15. 2025Accelerating Diffusion Large Language Models with SlowFast Sampling: The Three Golden Principlescited by 45
  16. 2025WeDLM: Reconciling Diffusion Language Models with Standard Causal Attention for Fast Inferencecited by 41
  17. 2025Plan for Speed: Dilated Scheduling for Masked Diffusion Language Modelscited by 34
  18. 2025A Survey on Parallel Text Generation: From Parallel Decoding to Diffusion Language Modelscited by 30
  19. 2025Roll the dice & look before you leap: Going beyond the creative limits of next-token predictioncited by 29
  20. 2025From Bits to Rounds: Parallel Decoding with Exploration for Diffusion Language Modelscited by 23
  21. 2025Improving Discrete Diffusion Unmasking Policies Beyond Explicit Reference Policiescited by 22
  22. 2025Climbing the Ladder of Reasoning: What LLMs Can-and Still Can't-Solve after SFT?cited by 21
  23. 2025Lookahead Unmasking Elicits Accurate Decoding in Diffusion Language Modelscited by 17
  24. 2025LoPA: Scaling dLLM Inference via Lookahead Parallel Decodingcited by 17
  25. 2026Improving Diffusion Language Model Decoding through Joint Search in Generation Order and Token Spacecited by 10
  26. 2025On Powerful Ways to Generate: Autoregression, Diffusion, and Beyondcited by 6
  27. 2026Discrete Diffusion Models Exploit Asymmetry to Solve Lookahead Planning Taskscited by 4

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026From Interface to Inference: Eliciting Any-Order Inference from Any-Order Modelscited by 1
  2. 2026Focus on the Core: Empowering Diffusion Large Language Models by Self-Contrastcited by 0
  3. 2026Escaping Confidence Trap: Evolutionary Decoding for Mathematical Reasoning in Diffusion LLMscited by 0
  4. 2026Information-Guided Frontier Decoding: Contextual Utility-Driven Commitment in dMLLMscited by 0
  5. 2026Low-Confidence Remasking Traps Flexibility: Realizing Arbitrary-Order Potential for Diverse Rollouts in Diffusion LLMscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Re-evaluating Confidence Remasking in Masked Diffusion Language Models
  2. 2026NAVIRA: Decoupled Stochastic Remasking for Masked Diffusion Language Models
  3. 2026CoDR: Training-Free Confidence-Drift Remasking for Diffusion Language Models
  4. 2026Confidence-Based Decoding is Provably Efficient for Diffusion Language Models
  5. 2026A Tale of Two Temperatures: Simple, Efficient, and Diverse Sampling from Diffusion Language Models
  6. 2025Beyond Confidence: Adaptive and Coherent Decoding for Diffusion Language Models
  7. 2026Roll Out and Roll Back: Diffusion LLMs are Their Own Efficiency Teachers
  8. 2026Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration