What it is
Small language models have far fewer parameters than large foundation models and are optimized for speed, cost, and narrow tasks rather than broad reasoning. They can run on smaller infrastructure and are often fine-tuned for jobs like classification or intent detection. Many agentic systems route simple tasks to SLMs and complex ones to larger models.
Why it matters
SLMs cut inference cost and latency, letting mid-market companies run AI at scale without runaway compute bills. Instead of sending every request through a large foundation model, teams can route routine work—like sorting cases or flagging intent—to a cheaper, faster SLM. That stretches AI budgets further and keeps agent responses quick as Agentforce scales across sales and service.
Key components
- Parameter-efficient architecture
- Task-specific fine-tuning
- Low-latency inference
- Model routing
How it connects
Agentforce and LLM gateways can route specific agent actions to SLMs for cost efficiency while reserving foundation models for complex reasoning.
Good to know
Not every agent task needs a giant model—matching task complexity to model size is a fast way to cut AI costs.
Related terms
Foundation Model
A large AI model trained on broad data at massive scale that serves as a general-purpose substrate other applications fine-tune or prompt against. The category that includes GPT, Claude, Gemini, Grok, Llama, and the rest of the frontier model lineup.
Inference Chips
Specialized silicon designed to run trained AI models (the "inference" step) rather than train them — typically optimized for low latency, high tokens-per-second, and lower cost-per-token than the general-purpose GPUs used for training.
LLM Gateway
A unified proxy in front of multiple LLM providers that captures every call, enforces policy, and lets a single application talk to Anthropic, OpenAI, xAI, Gemini, and local models through one interface.
LLM Cost Attribution
The practice of tying every LLM call back to the task, agent, process, or skill that triggered it — across every vendor — so AI spend can be measured against outcomes, not just tokens.
