"Harnessing Multiple Large Language Models: A Survey on LLM Ensemble" (ArXiv 2025)

Zhijun Chen, Jingzheng Li, Pengpeng Chen, Zhuoran Li, Kai Sun, Yuankai Luo, Qianren Mao, Dingqi Yang, Hailong Sun, Philip S. Yu

🔥🔥🔥 This is a collection of papers on LLM Ensemble.

[Always] [Add your paper in this repo] Thank you to all the papers that have cited our survey! We will add all related citing papers to this GitHub repo, in a timely manner, to help increase the visibility of your contributions.

[Always] [Maintain] We will make this list updated frequently (at least until 12/31/2025)!
If you found any error or any missed/new paper, please don't hesitate to contact us or Pull requests.

• Contents
  • 1 LLM Ensemble and Taxonomy
    • 1.1 LLM Ensemble
    • 1.2 Taxonomy
  • 2 Papers
    • 2.1 Ensemble Before Inference
    • 2.2 Ensemble During Inference
    • 2.3 Ensemble After Inference
    • 2.4 Others: Benchmarks, Applications and Related Surveys
  • 3 Others: Some pubilic implementations of the LLM Ensemble methods
  • 4 Others: Some other related interesting papers
  • 5 Summarization
  • 6 Citation

Paper Abstract:

LLM Ensemble---which involves the comprehensive use of multiple large language models (LLMs), each aimed at handling user queries during the downstream inference, to benefit from their individual strengths---has gained substantial attention recently. The widespread availability of LLMs, coupled with their varying strengths and out-of-the-box usability, has profoundly advanced the field of LLM Ensemble. This paper presents the first systematic review of recent developments in LLM Ensemble. First, we introduce our taxonomy of LLM Ensemble and discuss several related research problems. Then, we provide a more in-depth classification of methods under the broad categories of ``ensemble-before-inference, ensemble-during-inference, ensemble-after-inference'', and review all relevant methods. Finally, we introduce related benchmarks and applications, summarize existing studies, and suggest several future research directions. A curated list of papers on LLM Ensemble is available at https://github.com/junchenzhi/Awesome-LLM-Ensemble.

Figure 1: Illustration of LLM Ensemble Taxonomy. (Note that for (b) ensemble-during-inference paradigm, there is also a process-level ensemble approach that we have not represented in the figure, mainly because that this approach is instantiated by a single method.)

Figure 2: Taxonomy of All LLM Ensemble Methods. (Please note that this figure may not be fully updated to include all the papers listed below.)

• (a) Ensemble before inference.
  In essence, this approach employs a routing algorithm prior to LLM inference to allocate a specific query to the most suitable model, allowing the selected model that is specialized for the query and typically more cost-efficient inference to perform the task. Existing methods can be classified into two categories, depending on whether the router necessitates the use of pre-customized data for pre-training:

  • (a1) Pre-training router;
  • (a2) Non pre-training router.

• (b) Ensemble during inference.
  As the most granular form of ensemble among the three broad categories, this type of approach encompasses:

  • (b1) Token-level ensemble methods, which integrate the token-level outputs of multiple models at the finest granularity of decoding;
  • (b2) Span-level ensemble methods, which conduct ensemble at the level of a sequence fragment (e.g., a span of four words);
  • (b3) Process-level ensemble methods, which select the optimal reasoning process step-by-step within the reasoning chain for a given complex reasoning task. Note that for these ensemble-during-inference methods, the aggregated text segments will be concatenated with the previous text and fed again to models.

• (c) Ensemble after inference.
  These methods can be classified into two categories:

  • (c1) Non cascade methods, which perform ensemble using multiple complete responses contributed from all LLM candidates;
  • (c2) Cascade methods, which consider both performance and inference costs, progressively reasoning through a chain of LLM candidates largely sorted by model size to find the most suitable inference response.

Figure 3: Summary analysis of the key attributes of ensemble-before-inference methods. (Please note that this table may not be fully updated to include all the papers listed below.)

• LLM Routing with Benchmark Datasets. (2023)
            Name: -, Code: -

• RouteLLM: Learning to Route LLMs with Preference Data. (2024)
            Name: RouteLLM,

• Hybrid LLM: Cost-Efficient and Quality-Aware Query Routing. (2024)
            Name: Hybrid-LLM,

• LLM Bandit: Cost-Efficient LLM Generation via Preference-Conditioned Dynamic Routing. (2025)
            Name: -, Code: -

• Harnessing the Power of Multiple Minds: Lessons Learned from LLM Routing. (2024)
            Name: -,

• MetaLLM: A High-performant and Cost-efficient Dynamic Framework for Wrapping LLMs. (2024)
            Name: MetaLLM,

• SelectLLM: Query-Aware Efficient Selection Algorithm for Large Language Models. (2024)
            Name: SelectLLM, Code: -

• Bench-CoE: a Framework for Collaboration of Experts from Benchmark. (2024)
            Name: Bench-CoE,

• Routing to the Expert: Efficient Reward-guided Ensemble of Large Language Models. (2023)
            Name: ZOOTER, Code: -

• TensorOpera Router: A Multi-Model Router for Efficient LLM Inference. (2024)
            Name: TO-Router, Code: -

• Query Routing for Homogeneous Tools: An Instantiation in the RAG Scenario. (2024)
            Name: HomoRouter, Code: -

• Fly-Swat or Cannon? Cost-Effective Language Model Choice via Meta-Modeling. (2023)
            Name: FORC,

• Routoo: Learning to Route to Large Language Models Effectively. (2024)
            Name: Routoo, Code: -

• (Newly added paper, March 2025:) RouterDC: Query-Based Router by Dual Contrastive Learning for Assembling Large Language Models. (2024)
            Name: RouterDC,

• (Newly added paper, May 2025:) Rethinking Predictive Modeling for LLM Routing: When Simple kNN Beats Complex Learned Routers. (2025)
            Name: -, Code: -

• (Newly added paper, May 2025:) InferenceDynamics: Efficient Routing Across LLMs through Structured Capability and Knowledge Profiling. (2025)
            Name: InferenceDynamics, Code: -

• (Newly added paper, May 2025:) CO-OPTIMIZING RECOMMENDATION AND EVALUATION FOR LLM SELECTION. (2025)
            Name: RELM, Code: -

• (Newly added paper, May 2025:) Route to Reason: Adaptive Routing for LLM and Reasoning Strategy Selection. (2025)
            Name: RTR,

• (Newly added paper, June 2025:) RadialRouter: Structured Representation for Efficient and Robust Large Language Models Routing. (2025)
            Name: RadialRouter, Code: -

• (Newly added paper, June 2025:) TAGROUTER: Learning Route to LLMs through Tags for Open-Domain Text Generation Tasks. (2025)
            Name: TagRouter, Code: -

• PickLLM: Context-Aware RL-Assisted Large Language Model Routing. (2024)
            Name: PickLLM, Code: -

• Eagle: Efficient Training-Free Router for Multi-LLM Inference. (2024)
            Name: Eagle, Code: -

• Blending Is All You Need: Cheaper, Better Alternative to Trillion-Parameters LLM. (2024)
            Name: Blending, Code: -

• (Newly added paper, June 2025:) The Avengers: A Simple Recipe for Uniting Smaller Language Models to Challenge Proprietary Giants. (2025)
            Name: Avengers,

Figure 4: Summary analysis of the key attributes of ensemble-during-inference methods. (Please note that this table may not be fully updated to include all the papers listed below.)

• Breaking the Ceiling of the LLM Community by Treating Token Generation as a Classification for Ensembling. (2024)
            Name: GaC,

• Ensemble Learning for Heterogeneous Large Language Models with Deep Parallel Collaboration. (2024)
            Name: DeePEn,

• Bridging the Gap between Different Vocabularies for LLM Ensemble. (2024)
            Name: EVA,

• Determine-Then-Ensemble: Necessity of Top-k Union for Large Language Model Ensembling. (2024)
            Name: UniTe, Code: -

• Pack of LLMs: Model Fusion at Test-Time via Perplexity Optimization. (2024)
            Name: PackLLM,

• Purifying large language models by ensembling a small language model. (2024)
            Name: -, Code: -

• CITER: Collaborative Inference for Efficient Large Language Model Decoding with Token-Level Routing. (2025)
            Name: CITER,

• (Newly added paper, April 2025:) An Expert is Worth One Token: Synergizing Multiple Expert LLMs as Generalist via Expert Token Routing. (2024)
            Name: ETR,

• (Newly added paper, April 2025:) Speculative Ensemble: Fast Large Language Model Ensemble via Speculation. (2025)
            Name: Speculative Ensemble,

• Cool-Fusion: Fuse Large Language Models without Training. (2024)
            Name: Cool-Fusion, Code: -

• Hit the Sweet Spot! Span-Level Ensemble for Large Language Models. (2024)
            Name: SweetSpan, Code: -

• SpecFuse: Ensembling Large Language Models via Next-Segment Prediction. (2024)
            Name: SpecFuse, Code: -

• (Newly added paper, June 2025:) RLAE: Reinforcement Learning-Assisted Ensemble for LLMs. (2025)
            Name: RLAE, Code: -

• Ensembling Large Language Models with Process Reward-Guided Tree Search for Better Complex Reasoning. (2024)
            Name: LE-MCTS, Code: -

Figure 5: Summary analysis of the key attributes of ensemble-after-inference methods. (Please note that this table may not be fully updated to include all the papers listed below.)

• More Agents Is All You Need. (2024)
            Name: Agent-Forest,

• Smoothie: Label Free Language Model Routing. (2024)
            Name: Smoothie,

• Getting MoRE out of Mixture of Language Model Reasoning Experts. (2023)
            Name: MoRE,

• LLM-Blender: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion. (2023)
            Name: LLM-Blender,

• LLM-TOPLA: Efficient LLM Ensemble by Maximising Diversity. (2024)
            Name: LLM-TOPLA,

• URG: A Unified Ranking and Generation Method for Ensembling Language Models. (2024)
            Name: URG, Code: -

• (Newly added paper, April 2025:) DFPE: A Diverse Fingerprint Ensemble for Enhancing LLM Performance. (2025)
            Name: DFPE,

• (Newly added paper, April 2025:) Two Heads are Better than One: Zero-shot Cognitive Reasoning via Multi-LLM Knowledge Fusion. (2024)
            Name: MLKF,

• (Newly added paper, April 2025:) Symbolic Mixture-of-Experts: Adaptive Skill-based Routing for Heterogeneous Reasoning. (2025)
            Name: Symbolic-MoE,

• (Newly added paper, April 2025:) BALANCING ACT: DIVERSITY AND CONSISTENCY IN LARGE LANGUAGE MODEL ENSEMBLES. (2025)
            Name: DMoA, Code: -

• (Newly added paper, June 2025:) EL4NER: Ensemble Learning for Named Entity Recognition via Multiple Small-Parameter Large Language Models. (2025)
            Name: EL4NER, Code: -

• (Newly added paper, September 2025:) LENS: Learning Ensemble Confidence from Neural States for Multi-LLM Answer Integration. (2025)
            Name: EL4NER, Code: -

• EcoAssistant: Using LLM Assistant More Affordably and Accurately. (2023)
            Name: EcoAssistant,

• Large Language Model Cascades with Mixture of Thoughts Representations for Cost-efficient Reasoning. (2023)
            Name: -,

• Model Cascading: Towards Jointly Improving Efficiency and Accuracy of NLP Systems. (2022)
            Name: Model Cascading, Code: -

• Cache & Distil: Optimising API Calls to Large Language Models. (2023)
            Name: neural caching,

• A Unified Approach to Routing and Cascading for LLMs. (2024)
            Name: Cascade Routing,

• When Does Confidence-Based Cascade Deferral Suffice? (2023)
            Name: -, Code: -

• FrugalGPT: How to Use Large Language Models While Reducing Cost and Improving Performance. (2023)
            Name: FrugalGPT, Code: -

• Language Model Cascades: Token-level uncertainty and beyond. (2024)
            Name: FrugalGPT, Code: -

• AutoMix: Automatically Mixing Language Models. (2023)
            Name: AutoMix,

• Dynamic Ensemble Reasoning for LLM Experts. (2024)
            Name: DER, Code: -

• (Newly added paper, April 2025:) EMAFusionTM: A SELF-OPTIMIZING SYSTEM FOR SEAMLESS LLM SELECTION AND INTEGRATION. (2025)
            Name: EMAFusionTM, Code: -

• (Newly added paper, June 2025:) Do We Truly Need So Many Samples? Multi-LLM Repeated Sampling Efficiently Scales Test-Time Compute. (2025)
            Name: ModelSwitch,

• LLM-BLENDER: Ensembling Large Language Models with Pairwise Ranking and Generative Fusion. (2023)
            Name: MixInstruct, Evaluation Goal: Performance,

• RouterBench: A Benchmark for Multi-LLM Routing System. (2024)
            Name: RouterBench, Evaluation Goal: Performance and cost,

• (Newly added paper, April 2025:) RouterEval: A Comprehensive Benchmark for Routing LLMs to Explore Model-level Scaling Up in LLMs. (2025)
            Name: Speculative Ensemble,

Beyond the methods presented before, the concept of LLM Ensemble has found applications in a variety of more specialized tasks and domains. Here we give some examples:

• Ensemble-Instruct: Generating Instruction-Tuning Data with a Heterogeneous Mixture of LMs. (2023)
            Name: Ensemble-Instruct, Task: Instruction-Tuning Data Generation,

• Bayesian Calibration of Win Rate Estimation with LLM Evaluators. (2024)
            Name: BWRS, Bayesian Dawid-Skene, Task: Win Rate Estimation,

• PromptMind Team at MEDIQA-CORR 2024: Improving Clinical Text Correction with Error Categorization and LLM Ensembles. (2024)
            Name: -, Task: SQL generation, Code: -

• LLM-Ensemble: Optimal Large Language Model Ensemble Method for E-commerce Product Attribute Value Extraction. (2024)
            Name: -, Task: Product Attribute Value Extraction, Code: -

• (Newly added paper, April 2025:) FuseGen: PLM Fusion for Data-generation based Zero-shot Learning. (2024)
            Name: FuseGen, Task: Data-generation,

• (Newly added paper, April 2025:) On Preserving the Knowledge of Long Clinical Texts. (2023)
            Name: -, Task: Prediction tasks on long clinical notes, Code: -

• (Newly added paper, April 2025:) Consensus Entropy: Harnessing Multi-VLM Agreement for Self-Verifying and Self-Improving OCR. (2025)
            Name: Consensus Entropy (CE), Task: Optical Character Recognition, Code: -

• (Newly added paper, June 2025:) Beyond Monoliths: Expert Orchestration for More Capable, Democratic, and Safe Large Language Models. (2025)
            Name: Expert Orchestration, Task: -, Code: -

• (Newly added paper, May 2025:) Model Merging in LLMs, MLLMs, and Beyond: Methods, Theories, Applications and Opportunities. (2024)
            GitHub:

• (Newly added paper, May 2025:) Merge, Ensemble, and Cooperate! A Survey on Collaborative Strategies in the Era of Large Language Models. (2024)
            GitHub: -

• (Newly added paper, May 2025:) A Survey on Collaborative Mechanisms Between Large and Small Language Models. (2025)
            GitHub: -

• (Newly added paper, May 2025:) A comprehensive review on ensemble deep learning: Opportunities and challenges. (2023)
            GitHub: -

• (Newly added paper, May 2025:) Doing More with Less – Implementing Routing Strategies in Large Language Model-Based Systems: An Extended Survey. (2025)
            GitHub: -

• (Newly added paper, May 2025:) A Survey on Model MoErging: Recycling and Routing Among Specialized Experts for Collaborative Learning. (2024)
            GitHub: -

• (Newly added paper, May 2025:) Deep Model Fusion: A Survey. (2023)
            GitHub: -

• (Newly added paper, June 2025:) Towards Efficient Multi-LLM Inference: Characterization and Analysis of LLM Routing and Hierarchical Techniques. (2025)
            GitHub: -

• (Newly added paper, July 2025:) Toward Edge General Intelligence with Multiple-Large Language Model (Multi-LLM): Architecture, Trust, and Orchestration. (2025)
            GitHub: -

• (Newly added, May 2025:) Ensemble-Hub. (2025)
            GitHub:

Here we briefly list some related papers, which are either discovered by us or suggested by the authors to this repository. They mainly focus on LLM Collaboration.

• (Newly added paper, June 2025:) If Multi-Agent Debate is the Answer, What is the Question? (2025)
            Name: Heter-MAD, Code: -

• (Newly added paper, June 2025:) Nature-Inspired Population-Based Evolution of Large Language Models (2025)
            Name: GENOME,

Figure 6: Summary analysis of the key attributes of LLM Ensemble approaches.

@article{chen2025harnessing,
  title={Harnessing Multiple Large Language Models: A Survey on LLM Ensemble},
  author={Chen, Zhijun and Li, Jingzheng and Chen, Pengpeng and Li, Zhuoran and Sun, Kai and Luo, Yuankai and Mao, Qianren and Yang, Dingqi and Sun, Hailong and Yu, Philip S},
  journal={arXiv preprint arXiv:2502.18036},
  year={2025}
}