You commented “FOUR”: here it is

Which Claude to Use: Model Selection Cheat Sheet

Get the one-page breakdown that tells you exactly which Claude model fits any task, so you stop overpaying or underpowering.

Free checklist · copy-paste ready · built by the BrainVaultAI team

Anthropic ships four Claude models at any given time. Most people use one model for everything. That's either burning money on Opus for simple tasks or getting underwhelming output from Haiku on tasks that need depth. The cheat sheet below maps task types to the right model. The decision logic is simple: cost, speed, and reasoning depth. Get the matrix in your head once and the right call becomes automatic. This applies whether you're using the API, Claude.ai, or Claude Code. In Claude Code specifically, you can override the model per-session or per-agent, this cheat sheet tells you when to bother. The one rule that overrides everything: if the output is going to a customer or a decision that costs real money, don't cut corners on the model.

Model selection cheat sheet

## Which Claude to use ### Haiku (fastest, cheapest)
Use for:
- Classifying, tagging, or routing inputs
- Summarizing structured content (receipts, logs, short docs)
- Generating short copy variations (subject lines, headlines)
- Worker agents in a multi-agent pipeline where the task is narrow
- Anything where you're making 100+ calls and the task is well-defined Do NOT use for:
- Anything requiring multi-step reasoning
- Code generation beyond boilerplate
- Strategic analysis
- Anything a customer will read ### Sonnet (best balance, use this 80% of the time)
Use for:
- Code generation and debugging
- Draft writing for customers
- Research synthesis across multiple sources
- Orchestrating multi-agent workflows
- Complex but well-scoped tasks
- Your main Claude Code sessions Default to Sonnet when in doubt. ### Opus (deepest reasoning, highest cost)
Use for:
- Architectural decisions with long-term consequences
- Evaluating complex trade-offs
- Writing that needs to be exceptional, investor memos, positioning docs
- Tasks where getting it wrong is expensive
- When Sonnet's output is missing something important Do NOT use Opus by default. Use it when Sonnet fails. ### The one override rule
If the output touches a customer or a real money decision: don't use Haiku. Use at minimum Sonnet.

Quick decision tree

## 30-second model decision Q1: Is this task narrow, well-defined, and repeated 50+ times?
→ YES: Haiku
→ NO: go to Q2 Q2: Does this require code, multi-step reasoning, or customer-facing output?
→ YES: Sonnet
→ NO: go to Q3 Q3: Is getting this wrong expensive or irreversible?
→ YES: Opus
→ NO: Sonnet Q4: Are you unsure?
→ Sonnet. Always the safe default. ## In Claude Code
Change model per session:
claude --model claude-opus-4-5 # one session
claude --model claude-haiku-4-5 # one session Or set in ~/.claude/settings.json for a persistent default:
{ "model": "claude-sonnet-4-6"
}

Want the editable version?

We'll send the editable version of this asset, plus a short series of our highest-leverage AI plays for operators. No spam. Unsubscribe in one click.

Free. No spam. Unsubscribe in one click. By submitting you agree to receive emails from BrainVaultAI.

More free AI guides in the full library · follow @brainVaultAI