mlsanityData to 14 Oct 2026

Refine Thought: A Test-Time Inference Method for Embedding Model Reasoning

arXiv 2511.13726 ↗First posted 14 Oct 2025cs.CL cs.AI cs.IRcited by 2

On the map: LLM reasoning → Knowledge- and tool-augmented reasoning

Abstract (arXiv)

We propose RT (Refine Thought), a method that can enhance the semantic reasoning ability of text embedding models. The method obtains the final semantic representation by running multiple forward passes of the text embedding model. Experiments show that RT achieves significant improvements on semantic reasoning tasks in BRIGHT and the person-job matching benchmark PJBenchmark, while maintaining consistent performance on general-purpose semantic understanding tasks such as C-MTEB. Our results indicate that RT is effective because it further activates the semantic reasoning ability learned during pretraining by decoder-only text embedding models (e.g., Qwen3-Embedding-8B). RT can be seen as a test-time inference method.

Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Placed in Knowledge- and tool-augmented reasoning, one of 17 directions in LLM reasoning. Builds on 19 of its 21 references that are in the map. First posted on arXiv 14 Oct 2025.

Builds on

19 of 21 references.

References · 19

  1. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  2. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,459
  3. 2023Let's Verify Step by Stepcited by 3,789
  4. 2024Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameterscited by 1,986
  5. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,261
  6. 2014Neural Turing Machinescited by 644
  7. 2022Towards Reasoning in Large Language Models: A Surveycited by 596
  8. 2023C-Pack: Packed Resources For General Chinese Embeddingscited by 578
  9. 2024A Survey on RAG Meeting LLMs: Towards Retrieval-Augmented Large Language Modelscited by 452
  10. 2024Rewarding Progress: Scaling Automated Process Verifiers for LLM Reasoningcited by 279
  11. 2024Chain of Thought Empowers Transformers to Solve Inherently Serial Problemscited by 260
  12. 2025Reasoning with Latent Thoughts: On the Power of Looped Transformerscited by 240
7 more
  1. 2024BRIGHT: A Realistic and Challenging Benchmark for Reasoning-Intensive Retrievalcited by 158
  2. 2025Agentic Large Language Models, a surveycited by 108
  3. 2025A Survey on Latent Reasoningcited by 81
  4. 2024Procedural Knowledge in Pretraining Drives Reasoning in Large Language Modelscited by 41
  5. 2025Reinforcement Pre-Trainingcited by 40
  6. 2025RLP: Reinforcement as a Pretraining Objectivecited by 25
  7. 2025Memory Tokens: Large Language Models Can Generate Reversible Sentence Embeddingscited by 6

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 2.

2 papers.

  1. 2026A Survey of Reasoning-Intensive Retrieval: Progress and Challengescited by 1
  2. 2026CRE-T1 Preview Technical Report: Beyond Contrastive Learning for Reasoning-Intensive Retrievalcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Exploring Reasoning-Infused Text Embedding with Large Language Models for Zero-Shot Dense Retrieval
  2. 2024RAR-b: Reasoning as Retrieval Benchmark
  3. 2025Efficient Post-Training Refinement of Latent Reasoning in Large Language Models
  4. 2026LaME: Learning to Think in Latent Space for Multimodal Embedding via Information Bottleneck
  5. 2025Reasoning Guided Embeddings: Leveraging MLLM Reasoning for Improved Multimodal Retrieval
  6. 2025Latent Reasoning via Sentence Embedding Prediction
  7. 2026Think When Needed: Adaptive Reasoning-Driven Multimodal Embeddings with a Dual-LoRA Architecture
  8. 2026Rotate2Think: Geometric Priming via Orthogonal Rotation to Improve Language Model Reasoning