Choosing the Right Model for the Job
Frontier vs. fast models, reasoning effort, cost/latency trade-offs, and a practical decision framework for picking a model per task.
Most people pick a model once and never think about it again. Some always use the biggest, slowest, most expensive model "to be safe" and wonder why their agent costs so much and feels sluggish. Others always use the cheapest and wonder why it keeps making mistakes. Both are leaving a lot on the table.
The model is the biggest single lever on an agent's quality, speed, and cost, and the right choice depends on the task, not on which model is "best". In this lesson you'll learn how model tiers differ, what reasoning effort controls, how to weigh quality against cost and latency, and a simple decision framework you can apply to any task. You'll finish by running the same task on three models side by side and seeing the trade-offs for yourself.
Model tiers: frontier, balanced, and fast
Every major provider offers a family of models at different sizes. Larger models are generally smarter and more reliable on hard tasks; smaller ones are faster and much cheaper. The tiers look roughly like this:
| Tier | Strengths | Trade-offs | Good for |
|---|---|---|---|
| Frontier (largest) | Best reasoning, most reliable on long, complex, ambiguous tasks | Slowest and most expensive | Hard coding, deep research, planning, high-stakes work |
| Balanced (mid-size) | Strong on most everyday work at a fraction of the cost | Can slip on the hardest problems | Daily coding agents, most business tasks, chat products |
| Fast (smallest) | Very fast and very cheap | Weaker on multi-step reasoning and subtle judgment | Classification, extraction, routing, high-volume and simple subtasks |
Here's how that maps onto Anthropic's current Claude models:
| Model | API name | Tier |
|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 | Most capable, priced above Opus, for the hardest long-horizon work |
| Claude Opus 5.5 | claude-opus-5-5 | Frontier |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | Balanced |
| Claude Haiku 5.5 | claude-haiku-5-5 | Fast, and many times cheaper than the larger models |
OpenAI and Google have similar families, and the same reasoning applies to them. Specific model names will change; new versions arrive every few months. The tiers and the trade-offs are what stay stable, so learn those.