Retrieval-augmented generation in 2026: overview
Retrieval-augmented generation (RAG) supplies a language model with relevant external information when it responds. This can make answers more current and traceable than relying on model training alone, but retrieval does not guarantee a correct answer: a system can find a relevant source and still misinterpret it. The field is moving beyond fixed vector search toward adaptive retrieval, graph structures and repeated search-and-reason cycles. Publication activity rose 235.7% in the last year, while patents in the 24-month window declined 44.4%.
Recent work targets harder evidence types and more selective retrieval. In October 2026, TabulaRAG addressed numerical and spreadsheet data; NaviRAG, presented in August 2026 by researchers including Tsinghua University, explored navigating knowledge graphs. July 2026 work such as SARAG focused on whether retrieved evidence is sufficient, while UA-RAG explored uncertainty-aware dynamic retrieval. Healthcare remains an active test bed: OsteoRAG appeared in September 2026, and a glaucoma reasoning study followed in August 2026. Evaluation, fairness and security are also gaining attention, including Visa Inc.’s RAG-test and research on retrieval poisoning.
- 1Beyond fixed search: Adaptive and iterative systems aim to decide what to retrieve, whether the evidence is sufficient and when to stop searching.
- 2Structure matters: Graph-based methods such as NaviRAG target relationships across knowledge, while TabulaRAG highlights the limits of text-first retrieval for spreadsheet data.
- 3Trust needs testing: Work on evaluation, fairness and retrieval poisoning reflects a shift from grounding answers in sources to checking whether the process supports reliable answers.
Top organizations working on retrieval-augmented generation
Ranked by Linknovate activity score · last 24 monthsThe activity score adds up the relevance of every patent, publication, grant and news item linking an organization to retrieval-augmented generation. Showing 7 of 3,000 organizations.
| # | Organization | Approach | Signals (24 mo) | Momentum | Why it matters |
|---|---|---|---|---|---|
| 1 | 
Google
Corporate · United States |
General | PAT 12 · PUB 31 NEWS 894 |
-38% | Google applies retrieval to ground AI-generated answers in information from its search and other knowledge sources, including through AI Overviews and its AI products. |
| 2 | 
Microsoft
Corporate · United States |
Dense vector retrieval | PAT 32 · PUB 32 NEWS 649 |
-29% | Microsoft has patented RAG methods that create representative synthetic chunks for indexing and use documented expertise to improve the context supplied to assistants. |
| 3 | 
Amazon
Corporate · United States |
General | PAT 3 · PUB 43 NEWS 585 |
-54% | Amazon has patented techniques for protecting source data from being copied into RAG-generated answers, including systems that operate across sharded data sources. |
| 4 | 
Nvidia
Corporate · United States |
Dense vector retrieval | PAT 19 · PUB 20 NEWS 556 |
-33% | NVIDIA has patented expert-aware RAG that uses author metadata to prioritize contextually relevant information, and researchers have explored high-performance RAG for scientific question answering. |
| 5 | 
Tsinghua University
Research · China |
Knowledge graph retrieval | PUB 185 · NEWS 12 | -7% | Tsinghua researchers are developing graph-based RAG, including NaviRAG, which learns to navigate knowledge graphs to retrieve information for language-model responses. |
| 6 | 
OpenAI
Corporate · United States |
General | PUB 1 · NEWS 467 | +16% | OpenAI develops language models and tools that can use retrieved external information to ground responses; the available RAG coverage discusses the approach rather than a specific recent deployment. |
| 7 | 
Anthropic
Startup · United States |
Dense vector retrieval | PUB 1 · NEWS 432 | -5% | Anthropic is an AI model developer whose systems can support retrieval-grounded applications; recent discussion of RAG focuses on corpus design and balancing dense with lexical retrieval. |
| 8 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 9 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
| 10 | ··Hidden organizationCompany | Approach | PAT 00 · PUB 00 GRANTS 0 · NEWS 00 | +00% | Available to registered users. |
Startups to watch and leading research centers
Startups to watch 67 total
AuquanIndiaAuquan builds financial-services analytics using knowledge graphs and NLP, and has patented domain-specific RAG query processing for financial analysis.
AILabAILab researchers have worked on multimodal document reranking and methods for attributing generated answers to retrieved context, supporting more transparent RAG systems.- AAIRIFranceAIRI-related research examines adaptive RAG, which selects retrieval based on the question, and methods for handling compressed context in long-context generation.
- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
- ··Hidden startupCountryAvailable to registered users, with funding and founding year.
Leading research institutions 2,366 total
- Tsinghua UniversityChinaTsinghua researchers are developing graph-based RAG, including NaviRAG, which learns to navigate knowledge graphs to retrieve information for language-model responses.
University of Chinese Academy of SciencesChinaRecently, researchers at the university developed mR2AG, combining retrieval and reflection for knowledge-based visual question answering, and RagMedge, which grounds personalized medical dialogue in dense retrieval.
Chinese Academy of Sciences CASChinaResearchers affiliated with the academy have studied RAG security, including retrieval poisoning, as well as applications that combine retrieval with large language models at the network edge.- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
- ··Hidden research institutionCountryAvailable to registered users, with their key researchers.
Where retrieval-augmented generation innovation happens
Share of documents with at least one organization from each country.
Open the interactive map with all 3,000 organizationsMain technical approaches
Last 24 months| Approach | What it is | Documents | Leading organizations |
|---|---|---|---|
| Dense vector retrieval | Dense vector retrieval finds semantically similar passages, but can miss exact terms, numerical details or relationships that matter to a query. | 2,156 |
V-Rag, Google, Microsoft |
| Lexical and hybrid search | Lexical and hybrid search combine exact-term matching with semantic retrieval, improving coverage at the cost of added ranking and tuning complexity. | 745 |
V-Rag, Architectureplus International, Api(+), Elastic Inc. |
| Knowledge graph retrieval | Knowledge graph retrieval follows entities and their relationships to assemble connected evidence, but depends on building and maintaining structured graphs. | 2,175 |
Google, Microsoft, Amazon |
| Agentic and iterative retrieval | Agentic and iterative retrieval lets a system refine queries and search again, but extra reasoning and retrieval steps can increase latency and complexity. | 623 |
V-Rag, Google, Architectureplus International, Api(+) |
| Multimodal retrieval | Multimodal retrieval searches across text and visual content, but aligning different formats and preserving their context remains difficult. | 366 |
V-Rag, Zhejiang University, Nanyang Technological University |
Emerging trends
Last 12 months vs previous 36 monthsTerms whose frequency in the last 12 months grew the most compared with the previous 36 months.
Funding and investment
Recent funding rounds
Reported funding totals include $49,200,272 across 67 grants since 2020. The $850,225,135,050 listed for 59 rounds also aggregates broad corporate and infrastructure financing, so it is not a clean measure of RAG investment.
Patent landscape
Top patent holders · last 24 monthsRecent developments
Latest news, duplicates removed- Oct 2026Your RAG agent is only as good as its last retrieval. As it ships a wrong answer with a confident tone, the damage isn't a bug report. It's a trust problem that follows your product into every conversation after it.…Excellent Webworld
- Sep 2026Modern Large Language Models (LLMs) have transformed enterprise software engineering and automated content generation. However, deploying base foundation models directly in production quickly exposes their fundamental…Kunal-Chowdhury.com - Latest Tech News, Product Reviews, Tips and Tricks on Windows and Android
- Aug 20261. What Is RAG and What Problem Does It Solve? Large language models are trained once on a massive dataset and then deployed. After training concludes, their knowledge is permanently fixed—a snapshot of everything in…C-Metric.com
- Aug 2026how to take a RAG system past one document and the answer has been the same for three years. Put every page of every file into a vector store, embed the question, and let similarity search pick the passages. On a demo…Towards Data Science - Medium
- Month 20XXHidden developmentAvailable to registered users with alerts enabled.
Frequently asked questions about retrieval-augmented generation
What companies are working on retrieval-augmented generation?
Google, Microsoft, Amazon and Nvidia are among the companies active in retrieval-augmented generation. The listed organizations work on areas including grounding answers in search and knowledge sources, retrieval methods and protection of source data. Dell also has RAG patents, including methods for complex documents. The company landscape spans both commercial developers and research organizations.
Which universities and research centers lead retrieval-augmented generation research?
Tsinghua University leads the listed research organizations with 197 documents in the 24-month window, followed by the University of Chinese Academy of Sciences with 149. The Chinese Academy of Sciences has 127, and Zhejiang University has 130. Recent work includes Tsinghua’s graph-based NaviRAG and research on medical dialogue, security and retrieval methods.
How much funding has retrieval-augmented generation received?
The reported grant total is $49,200,272 across 67 grants since 2020. A separate record lists $850,225,135,050 across 59 funding rounds, but this includes broad corporate and infrastructure financing and should not be treated as RAG-specific investment. The two figures measure different funding categories and have different limitations.
Who holds the most retrieval-augmented generation patents?
Dell holds the most listed RAG patents, with 56; Microsoft is next with 32. Oracle has 22, Tata Consultancy Services Ltd. has 19, and Nvidia has 18. These counts identify leading holders in the supplied patent-holder data, while the broader patent activity measure covers a separate time window.
Is retrieval-augmented generation patent activity growing?
No: RAG patent activity declined 44.4% in the 24-month window comparison, with 893 patents in the window. Publication activity moved in the opposite direction, rising 235.7% in the last year to 5,425 publications. The divergence suggests that recent research activity is expanding faster than the patent measure.
What are the latest developments in retrieval-augmented generation?
Recent developments include TabulaRAG for numerical and spreadsheet data in October 2026, and NaviRAG for navigating knowledge graphs in August 2026. July 2026 work introduced SARAG, which considers whether retrieved evidence is sufficient, and UA-RAG, which explores uncertainty-aware dynamic retrieval. Recent studies also address medical knowledge, evaluation and fairness.
What are the main approaches in retrieval-augmented generation and how do they differ?
The main approaches are dense vector retrieval, lexical and hybrid search, knowledge graph retrieval, agentic and iterative retrieval, and multimodal retrieval. Vector search finds semantically similar passages; lexical search adds exact-term matching; graph methods use relationships; agentic methods refine searches; and multimodal systems retrieve across formats. Each approach trades retrieval coverage and structure against system complexity.
Which countries lead retrieval-augmented generation?
The United States leads with 10,692 documents, or 49.0% of the listed total, followed by China with 4,157 documents and 19.0%. The United Kingdom has 2,141 documents, while Germany has 1,900 and India has 1,866. These figures describe the geographic distribution of documents in the landscape.
What are the applications of retrieval-augmented generation?
Enterprise knowledge and customer support is the largest listed application, with 2,221 documents. Other areas include software development with 926, healthcare with 855, legal and compliance with 736, scientific research with 590, and education with 252. Recent studies also explore medical decision support, spreadsheet question answering and interactive video understanding.