Research brief for Gemini long-context

“research the current state of small language models for me”

CONTEXT

You are preparing a research brief on the current state of small language models (SLMs). The brief will be read by a product team at a SaaS company that is considering deploying an SLM for on-device or low-latency features.

Scope:
- "Small" = under ~8B parameters, suitable for consumer hardware or cheap edge deployment.
- Cover open-weights families (Llama, Gemma, Phi, Qwen, Mistral) and closed APIs only insofar as they compare on cost / latency.
- Focus on the last 12 months. Earlier models only as historical context.

What we already know (do not re-explain):
- The general transformer architecture.
- That bigger models are usually more capable.

What we need from you:
- A clear, plainly-written brief, not a list of papers.
- Concrete comparisons (cost, latency, quality on a public benchmark) where the numbers are publicly reported.
- Honest gaps: where the data is missing or unreliable, say so.

TASK

Based on the context above, produce the brief. Use this structure exactly:

1. One-paragraph executive summary (≤ 80 words).
2. The current best-in-class SLMs (3–5 models) with a one-line "why this one".
3. Cost / latency / quality comparison table (markdown).
4. Three use-cases where SLMs already beat frontier models in practice.
5. Three known limitations or open problems.
6. Recommended next steps for a product team evaluating SLMs.

Stay grounded in the context above. If you must invent a number, label it [estimate].