Business Strategy & ROIRole-Specific

Opus, Sonnet, or Haiku: which Claude model should your team use?

Claude has three model tiers. Here is which one to use for what — and why defaulting to the most powerful one is usually a mistake.

5 min read·Claude Plans

Claude comes in three models: Opus (most powerful), Sonnet (balanced), and Haiku (fastest and cheapest). Most teams default to whatever sounds best — usually Opus. This wastes money and often produces slower results without meaningful quality improvement.

Here is how to actually choose.

What each model is good at

Haiku — the workhorse. Fast responses, lowest cost. Use for:

Simple Q&A, classification, extraction
Reformatting text (turning bullet points into paragraphs, adjusting tone)
Sorting, categorising, and tagging
Quick drafts where speed matters more than nuance
High-volume tasks where cost scales linearly

Haiku handles 70-80% of what most teams use Claude for. If the task has a clear, straightforward answer, Haiku gets there.

Sonnet — the daily driver. Good at most things. Use for:

Content drafting (emails, blog posts, reports)
Summarising documents and conversations
Analysis that requires some judgment
Code generation and review
Multi-step tasks that need coherent reasoning

Sonnet is where most teams should spend most of their time. It handles complex work well, responds quickly, and costs significantly less than Opus.

Opus — the heavy lifter. Use when the problem is genuinely hard:

Complex analysis with ambiguous inputs
Tasks requiring nuanced judgment (evaluating strategy, reviewing arguments)
Long documents where maintaining coherence over 50+ pages matters
Difficult reasoning, math, or logic problems
When Sonnet's output is not good enough — use Opus as an upgrade, not a default

The mistake teams make

Defaulting to Opus for everything. It is like taking a taxi for every trip when most of them are a five-minute walk. You arrive at the same place, just slower and more expensively.

The right approach: Start with Sonnet. If the output quality isn't good enough for a specific use case, upgrade to Opus for that use case. Keep everything else on Sonnet. Route truly simple tasks (extraction, reformatting, classification) to Haiku.

On Claude.ai, you select the model in the model picker at the top of each conversation. On the API, you specify the model per request — which means you can route different types of tasks to different models programmatically.

For team admins: model strategy

If you're managing a team on Claude:

Set Sonnet as the default. Most people don't need to think about model selection. Sonnet handles their work.
Educate on when to upgrade. If someone is doing deep analysis or getting mediocre outputs, switching to Opus for that specific task is the right move.
Watch usage patterns. If your usage limits are getting hit, check whether people are using Opus for tasks Sonnet handles equally well. Switching to Sonnet for routine work extends your limits significantly.

On API pricing (for teams building with Claude)

The cost difference is substantial:

Haiku: ~$0.25 / million input tokens, $1.25 / million output tokens
Sonnet: ~$3 / million input tokens, $15 / million output tokens
Opus: ~$15 / million input tokens, $75 / million output tokens

Opus costs 5x more than Sonnet and 60x more than Haiku. For a production application processing thousands of requests, model routing — sending simple requests to Haiku and complex ones to Sonnet or Opus — is one of the highest-leverage cost optimisations you can make.

The simple rule