Living systematic review

Retrieval-augmented generation

Grounding LLM generation in retrieved documents — retrieval, reranking, graph/recursive indexing, self-correction, and robustness of the retrieve-then-generate pipeline.

664 papers1,433 critique receipts6,345 benchmark resultsupdated Jun 20, 2026

Most-superseded baselines

Ranked by how many distinct papers critique or beat each method — the standard baselines newer work routinely measures against.

  1. 1
    RAG

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks

    69 critique · 79 beaten on benchmarks

  2. 2
    GraphRAGin RAG

    A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models

    38 critique · 31 beaten on benchmarks

  3. 3
    Self-RAG

    Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection

    26 critique · 40 beaten on benchmarks

  4. 4
    LightRAGin RAG

    15 critique · 24 beaten on benchmarks

  5. 5
    RAPTORin RAG

    RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval

    7 critique · 21 beaten on benchmarks

  6. 6
    FLAREin Self-RAG

    Active Retrieval Augmented Generation

    11 critique · 14 beaten on benchmarks

  7. 7
    PoisonedRAG

    PoisonedRAG: Knowledge Corruption Attacks to Retrieval-Augmented Generation of Large Language Models

    12 critique · 13 beaten on benchmarks

  8. 8
    HippoRAG 2in RAG

    From RAG to Memory: Non-Parametric Continual Learning for Large Language Models

    1 critique · 23 beaten on benchmarks

  9. 9
    IRCoTin RAG

    4 critique · 19 beaten on benchmarks

  10. 10
    BM25

    6 critique · 16 beaten on benchmarks

  11. 11
    CRAGin Self-RAG

    Corrective Retrieval Augmented Generation

    11 critique · 11 beaten on benchmarks

  12. 12
    ReActin RAG

    ReAct: Synergizing Reasoning and Acting in Language Models

    6 critique · 15 beaten on benchmarks

The competition

Methods that fight on the same benchmarks cluster into distinct sub-problems.

RAG369 methods

RAG · GraphRAG · LightRAG · RAPTOR · HippoRAG 2 · IRCoT

Self-RAG245 methods

Self-RAG · FLARE · CRAG · Adaptive-RAG · Iter-RetGen · DRAGIN

BM25123 methods

BM25 · DPR · ColBERT · RankRAG · Contriever · KG²RAG

RECOMP97 methods

RECOMP · LongLLMLingua · LLMLingua · CompAct · ICAE · xRAG

MMGraphRAG112 methods

MMGraphRAG · Video-RAG · VisRAG · RAG-Anything · KG-RAG · MiniCPM (OCR)

RobustRAG62 methods

RobustRAG · TrustRAG · Astute RAG · ReliabilityRAG · PPL · PromptGuard

RetRobust60 methods

RetRobust · RAFT · RAAT · Ret-Robust · LLM-Embedder · PA-RAG

PoisonedRAG48 methods

PoisonedRAG · Joint-GCG · AgentPoison · BadChain · Context-Agnostic Attack · GARAG

AutoRAG30 methods

AutoRAG · ReARTeR · FastRAG · LlamaIndex · DeepRAG · RAG-Star

RAG-DDR17 methods

RAG-DDR · Emotion-LLaMA · PandaGPT · Video-LLaMA · Video-LLaMA 2 · VideoChat

The frontier

Recent methods not yet superseded in the knowledge base.