New Pleias-RAG models offer citation and grounding for smaller LLMs

Researchers have introduced the Pleias-RAG model family, featuring smaller reasoning models designed for retrieval-augmented generation (RAG), search, and source summarization. These models, Pleias-RAG-350m and Pleias-RAG-1B, are mid-trained on a synthetic dataset that emulates multilingual open-source retrieval. They offer native support for citation and grounding with literal quotes, alongside features like query routing and source reranking. The models demonstrate superior performance on RAG benchmarks compared to other small language models (SLMs) and are competitive with larger models like Qwen 2.5 7B and Llama-3.1 8B, while also maintaining consistent performance across European languages and ensuring systematic reference grounding. AI

IMPACT These models could enable more factually grounded and efficient AI applications on constrained infrastructure.

RANK_REASON The cluster is about a new research paper introducing a family of models with novel capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• 2WikiMultiHopQA
• arXiv
• Common Corpus
• Gemma 3-4B
• HotpotQA
• Llama-3.1:8b
• Pavel Chizhov
• Pleias-RAG
• Pleias-RAG-1B
• Pleias-RAG-350m
• Qwen 2.5 7B

• model release
• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →