Survey maps how large and small language models can work together
A research survey organizes the emerging field of LLM-SLM collaboration, describing five interaction mechanisms and the open problems that still stand in the way of practical deployment.
In plain words
Large language models are powerful but expensive to run, while small language models are cheap and fast but less capable. The survey's authors argue that combining the two can balance those trade-offs, especially on devices with limited resources. They catalog how such pairings can interact and what problems remain unsolved.
What was announced
The arXiv survey, published as version 2 of paper 2505.07460, frames collaboration between large and small language models as a way to pair the capability of big models with the efficiency of small ones. The authors organize the field around five interaction mechanisms: pipeline, routing, auxiliary, distillation, and fusion. They also identify enabling technologies and application scenarios driven by on-device needs such as low latency, privacy, personalization, and offline operation.
The survey does not present new experimental results; it is a literature overview. The authors acknowledge persistent challenges, including system overhead, consistency between models, robust task allocation, evaluation complexity, and security and privacy concerns. They point to future work on adaptive frameworks, deeper model fusion, and extension to multimodal and embodied AI. Because the source is only the paper's abstract, the full survey's methods, coverage, and conclusions could not be verified here.
What it does not show
- The evidence is limited to the paper's abstract; the full survey's methodology, scope of literature covered, and detailed findings were not available for verification.
- The survey describes the field and open challenges but reports no new experimental measurements or independent validation of the collaboration approaches it catalogs.
For security and IT teams
- Scope
- The source is an academic survey abstract covering LLM-SLM collaboration mechanisms and challenges; it announces no product, tool, or deployment guidance for practitioners.
Sources
Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 20 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.