What question did this study set out to answer?

This study aims to evaluate the performance of various embedding models in RAG systems to retrieve specific technical identifiers.

July 7, 2026Open Access

nomic-embed-text Outperforms Larger Embedding Models on Technical Infrastructure RAG: A Three-Phase Controlled Study

View Full Paper

Ask AI

Bookmark

Authors

Controlled study compares embedding architectures for exact content retrieval in enterprise domains, suggesting size isn't always key.

This study aims to evaluate the performance of various embedding models in RAG systems to retrieve specific technical identifiers.
Conducted a three-phase controlled study comparing nomic-embed-text, Jina Embeddings v3, and BGE-M3 on CPU-only hardware.
Evaluated performance on a real-world infrastructure corpus with 451 content chunks and 30 cross-domain questions.
Used six open-weight language models for evaluation, ensuring offline conditions without GPU acceleration.
Nomic-embed-text outperforms larger models by an average of 6–10 percentage points in retrieval quality.
Hybrid retrieval mechanisms resolve some identifier retrieval issues but struggle with corpus ambiguity.
Llama3.1:8b demonstrates consistent performance across embedding models, indicating potential resilience in larger LLMs.

Eswar Raman (2026) studied this question.

View Full Paper

Ask AI

Bookmark