SLMs are well-positioned for the agentic era because they use a narrow slice of LLM functionality for any single language model errand. LLMs are built to be powerful generalists, but most agents use only a very narrow subset of their capabilities. They typically parse commands, generate structured outputs such as JSON for tool calls, or produce summaries and answer contextualized questions. These tasks are repetitive (up to the differences in prompt payloads), predictable, and highly specialized—well within the scope of specialized SLMs. An LLM trained to handle open-domain conversations is o
How Small Language Models Are Key to Scalable Agentic AI | NVIDIA Technical BlogIf SLMs have clear advantages, why do most agents still rely so heavily on LLMs? We hypothesize that the barriers are perception-based or caused by organizational culture rather than technical limitations. Shifting to SLM-enabled architectures requires an intentional mindset change. SLM research uses generalist benchmarks, even though agentic workloads demand different evaluation metrics. Plus, LLMs often dominate the headlines. As the cost savings and reliability of SLM-enabled systems become undeniable, momentum will shift. The transition could mirror past shifts in computing, such as the mo
How Small Language Models Are Key to Scalable Agentic AI | NVIDIA Technical Blog… Please reach out to us if you have any ideas or feedback. { "model": "MiniMaxAI/MiniMax-M2.5", "messages": ... , "tools": ... , "nvext": { "agent hints": { "osl": 256, "speculative prefill": true, "priority": 10 }, "cache control": { "type": "ephemeral", "ttl": "1h" } } } The agent hints fields: pr… …
… The trace makes it clear that agent token consumption is shaped as much by agentic system behavior as by the nature of the tasks. The primary agent accumulates input context quickly when it is not delegating or compacting context, which causes cache-read input token costs to recur every turn. …
… Enterprises looking to control costs, improve efficiency, and scale responsibly can begin experimenting with heterogeneous systems today. Conclusion: the heterogeneous system advantage The demand for agentic AI systems is rapidly evolving. …
… Nemotron 3 Nano 30B A3B Nemotron 3 Nano offers 4x faster throughput compared to Nemotron 2 Nano Leading accuracy for coding, reasoning, math and long context tasks Perfect for agents that need to deliver highest accuracy and efficiency for targeted tasks Nemotron 3 Nano Omni 30B A3B Single model fo… …
… From reliable retrieval to production-ready AI agents Once retrieval is stabilized, AI agents become more reliable because they operate on grounded context instead of improvisation. …
… These systems rely on agentic long‑term memory for context that persists across turns, tools, and sessions so agents can build on prior reasoning instead of starting from scratch on every request. …
… It governs how the agent executes, what the agent can see and do, and where inference goes. OpenShell enables claws to run in isolated sandboxes, giving you fine-grained control over your privacy and security while letting you benefit from the agents’ productivity. …
… With this, agents can perceive and reason across visual, audio, and textual inputs within a single shared perception‑to‑action loop, improving convergence and reducing orchestration complexity and inference cost. …
… MoE layers scale effective parameter count without the cost of dense computation. Only a subset of experts activates per token, keeping latency low and throughput high—critical when many agents are running concurrently in a shared deployment. …
… This foundational layer sets the cost of compute‑as‑a‑service, enforces where data can reside, and controls which tenants can run which workloads in a shared environment. …