Our report shows how an orchestrated ensemble of specialist SLMs matched and often outperformed frontier LLMs.

Small Language Models (SLMs) are becoming an increasingly important part of the enterprise AI landscape. As organisations move beyond experimentation and start deploying AI into operational workflows, there is growing demand for models that are smaller, faster, more cost-effective, and easier to run in controlled environments.

This is particularly relevant in regulated sectors, where data residency, privacy, latency, explainability, and operational control are just as important as raw model capability. Historically, standalone SLMs have not been considered a viable alternative to LLMs, however at Connect we have demonstrated that there does not need to be a trade off between capability and control, the way you train and architect AI models can matter more than model size alone.

Our benchmark report tested that premise across 19 enterprise tasks in finance, healthcare and insurance. We took our proprietary SLM across three configurations — standalone, domain-tuned and orchestrated as an ensemble of specialists – and benchmarked across the top frontier models The results show that:

1. Compact local SLMs deliver enterprise grade domain performance,

2. Domain adaptation makes a compact SLM more capable within each target sector
3. Orchestrating specialised SLMs improve reliability, groundedness, and suitability for regulated enterprise deployment

The orchestrated approach increased overall performance from 72.6% to 85.4%, while retaining the option to deploy locally and keep sensitive customer data inside the organisation's environment.

72.6% → 85.4%

Same compact model performing radically better results through domain tuning and orchestration.

+30 points

Accuracy leapt from 56% to 86% - ahead of every vendor model tested.

100% sovereign

Locally deployed to keep your data secure and your costs predictable.

You wouldn't ask one contact centre agent to handle every type of interaction. Why ask one AI model to do it?