Guides
Which AI model should your agent use? A practical guide
Fast and cheap for triage, strong reasoning for research and code, reliable tool calling for everything. How to pick a model per job, and when to switch.
The right model for an AI agent depends on the job. For high-volume, simple work like inbox triage, use a fast, inexpensive model. For long reading, research, or code review, use a strong reasoning model with a large context window. For every agent, the model must call tools reliably. Because each Stormi bot can use a different model, you don't have to pick one for everything.
The four things that matter
1. Tool calling. An AI agent works by asking for tools: search email, update a record. A model that fumbles tool calls (wrong arguments, calling the same tool in circles) makes a bad agent no matter how smart it is. Stormi's model picker only shows models that call tools and read images.
2. Reasoning. Multi-step jobs, like "find leads who went quiet, read their history, and write each a personal note," reward models that plan well. Stronger models often finish in fewer steps.
3. Context length. How much the model can read at once. Research digests and long threads need more; triage needs little.
4. Cost and speed. Priced per token. Small models can be many times cheaper and faster than frontier ones. See how much AI agents cost.
A model for each job
| Job | What matters most | Kind of model |
|---|---|---|
| Inbox triage, sorting, tagging | Speed, cost | Small, fast model |
| Drafting replies | Writing quality, tone | Mid-size or strong writer |
| Meeting briefs | Reading a few sources, concise output | Mid-size |
| Research digests | Long context, careful reading | Strong model, large context |
| Sales follow-ups | Personalization, tone | Strong writer |
| Code review | Code understanding, reasoning | Top coding model |
| Coordinating other bots | Planning, tool calling | Strong reasoning model |
Model names change monthly, so we describe kinds rather than crown a winner. In Stormi, each model's spec sheet shows its price, context length, what it reads and writes, and whether it reasons.
How to test a model for your job
- Pick a real task you've done by hand, like last Monday's inbox.
- Run it on two models: a cheap one and a strong one.
- Compare the result and the cost in Settings › Usage, which breaks spend down by bot and model.
- Keep the cheapest one that's right. If the cheap one is right 9 times in 10 and approvals catch the tenth, that's often the best deal.
Or let OpenRouter choose
Pick Auto and OpenRouter chooses a model per request. It's a good default while you learn what a bot needs.
When to switch models
- The bot keeps calling the wrong tool or repeating itself: try a stronger model.
- Results are fine but it's slow or costly: try a smaller one.
- A new model ships that's better at your job: switch the one bot that benefits. Nothing else changes.
Frequently asked questions
What is the best LLM for AI agents?
Can each agent use a different model?
Does a bigger model always make a better agent?
How do I see which model is costing the most?
Keep reading
Guides
How a multi-agent AI team works: the Chief of Staff pattern
One bot with fifty tools wanders. A team of bots with one job each, plus a Chief of Staff that hands out the work, finishes. Here's how to set one up.
3 min read
Guides
How to get to inbox zero with AI (and stay there)
An AI agent can triage your inbox every morning, file what doesn't need you, and draft the replies. Here's the exact setup, the job to give it, and the rules that keep it safe.
3 min read
Guides
Turn the iPhone Action button into an AI that sees your screen
Hold the Action button over a video, a post, or an article and ask a bot about it. The answer appears right there. Here's the three-step Shortcuts setup.
2 min read
Hire your first bot today
Stormi is free to download. Bots start on our model account, or connect your own OpenRouter account.
Download on the App Store