MIRAGE: Multi-Perspective Creative Language Model Reasoning with Reinforcement Learning Guidance
On the map: LLM reasoning → Reasoning for agents, planning and code also fits: Reasoning benchmarks and evaluation
Abstract (arXiv)
Recent advances in Large Language Models (LLMs) have revolutionized artificial intelligence and how human interact with AIs. Despite impressive advancements, LLMs struggle with complex mathematical, scientific, and logical tasks. Inspired by human cognitive flexibility - our ability to dynamically switch mental perspectives - we propose MIRAGE (Multi-perspective Inference-time Reasoning via Agent-Guided Exploration), a novel inference-time creative thinking framework. MIRAGE includes a Selector that prioritizes effective conceptual perspectives (e.g., algebraic, probabilistic) and a Reasoner that sequentially solves tasks until a confident solution emerges, otherwise aggregating multiple perspectives. Tested on GSM8K, MATH500, MMLU-Pro, and Game-of-24 benchmarks, MIRAGE consistently outperforms methods like Chain-of-Thought and diverse prompting ensembles, significantly boosting accuracy with minimal inference overhead, providing a scalable solution for practical applications.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Reasoning for agents, planning and code, one of 17 directions in LLM reasoning. Builds on 11 of its 58 references that are in the map. First posted on arXiv 18 Sep 2026.
Builds on
11 of 58 references.
References · 11
- 2021Training Verifiers to Solve Math Word Problemscited by 10,109
- 2020Measuring Massive Multitask Language Understandingcited by 7,963
- 2013Efficient Estimation of Word Representations in Vector Spacecited by 4,375
- 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
- 2020Fourier Neural Operator for Parametric Partial Differential Equationscited by 2,908
- 2018Relational inductive biases, deep learning, and graph networkscited by 1,360
- 2023HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Facecited by 1,190
- 2022STaR: Bootstrapping Reasoning With Reasoningcited by 940
- 2024DeepSeek-V3 Technical Reportcited by 811
- 2024HDFlow: Enhancing LLM Complex Problem-Solving with Hybrid Thinking and Dynamic Workflowscited by 13
- 2025When is the partial map classifier a Sierpi'nski cone?cited by 1
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2024MIRAGE: Evaluating and Explaining Inductive Reasoning Process in Language Models
- 2026MIRAGE: Mobile Agents with Implicit Reasoning and Generative World Models
- 2025Reasoning Capabilities of Large Language Models on Dynamic Tasks
- 2024Can large language models explore in-context?
- 2025Large Language Models Think Too Fast To Explore Effectively
- 2025T1: Advancing Language Model Reasoning through Reinforcement Learning and Inference Scaling
- 2025Do Theory of Mind Benchmarks Need Explicit Human-like Reasoning in Language Models?
- 2025Large Reasoning Models are not thinking straight: on the unreliability of thinking trajectories