1Why Multiple Models Exist
Think of AI models like hiring for different roles — you wouldn't assign a senior director to answer routine emails, or ask an intern to develop your program strategy. Each model is optimized for a different balance of intelligence, speed, and cost, and those dimensions trade off against each other.
Use the smallest, cheapest model that can do the job well. Upgrade only when the task demands it — this rule alone will cut your AI costs in half.
2The Model Family
Claude comes in three models — Haiku, Sonnet, and Opus. They share the same strengths (writing, analysis, reasoning, multi-part instructions) and differ only in the balance of speed, intelligence, and cost.
| Model | Best For | Key Specs |
|---|---|---|
| Claude Haiku 4.5 | Quick questions, data formatting, simple classification, bulk repetitive tasks | Fastest. Lowest cost. |
| Claude Sonnet 4.5 | Reports, data analysis, meeting summaries, email drafting, most daily work | Balanced speed and quality. Your default. |
| Claude Opus 4.7 | Complex strategy, deep multi-step analysis, important deliverables | Deepest reasoning. Slower. Premium. |
Claude doesn't process video or audio natively and doesn't generate images. For those niche tasks, a purpose-built tool from another provider is the right call — you don't need to learn these in depth, just know they exist.
| Task | Tool to Use |
|---|---|
| Analyze a video or audio recording | Google Gemini (native video/audio) |
| Generate an image from text | A dedicated image model (e.g. GPT Image 1 Mini) |
| Transcribe audio to text only | A dedicated transcription model (e.g. Whisper) |
Need to summarize a recorded training session or meeting? Claude cannot process video or audio directly. Google Gemini handles both natively and is the standout option for that specific job.
Sonnet is your default for 80% of tasks. Switch to Haiku for simple tasks, Opus for high-stakes reasoning.
3Choosing by Task
The fastest way to pick a model is to match your task to a category.
| Your Task | Use This Model | Why |
|---|---|---|
| Quick lookup or short question | Haiku 4.5 | Fast and cheap, no reasoning required |
| Reformat data into a table | Haiku 4.5 | Pattern-matching — Haiku handles it perfectly |
| Summarize meeting notes (1–2 pages) | Sonnet 4.5 | Balanced quality and speed |
| Draft an email to your director | Sonnet 4.5 | Strong tone control, follows complex instructions |
| Analyze a spreadsheet or report | Sonnet 4.5 | Strong analytical capability at moderate cost |
| Complex multi-factor decision analysis | Opus 4.7 | Deep reasoning required, worth the premium |
| Process a very large document (300+ pages) | Sonnet 4.5 | Large context window handles it in one pass |
| Process a video or audio recording | Google Gemini | Claude doesn't read video or audio |
Full task reference table
| Write a performance review | Sonnet 4.5 | Nuanced writing with good structure |
| Write a business case or strategy doc | Opus 4.7 | Needs sustained coherence over long outputs |
| Analyze images, screenshots, diagrams | Sonnet 4.5 | Handles images and text together |
| Extract data from many short documents | Haiku 4.5 | Fast and cheap for repetitive extraction |
| Transcribe audio to text only | A dedicated transcription tool | Purpose-built for audio-to-text |
| Generate an image from text | A dedicated image tool | Purpose-built for image generation |
The five-question decision flow
1. Is the task simple and well-defined? (formatting, lookup, classification) -> Use Haiku 4.5 2. Is the task standard knowledge work? (writing, summarizing, drafting) -> Use Sonnet 4.5 — YOUR DEFAULT 3. Does the task require deep reasoning? (strategy, complex decisions) -> Use Opus 4.7 4. Does the task involve video or audio? -> Video/audio analysis: Google Gemini -> Transcription only: a dedicated transcription tool 5. Does the task involve generating images? -> Use a dedicated image tool
If you can't clearly answer yes to question 1 or 3–5, use Sonnet. It's the right call for the vast majority of daily knowledge work.
4Speed, Cost, and Common Mistakes
| Model | Relative Cost | Best Value For |
|---|---|---|
| Haiku 4.5 | 1x (baseline) | Simple, repetitive, high-volume tasks |
| Sonnet 4.5 | ~3x Haiku | Standard daily knowledge work (your default) |
| Opus 4.7 | ~5x Haiku | High-stakes reasoning only |
1 token = ¾ of a word (about 4 characters) 1 page = 500 tokens 10 pages = 5,000 tokens 1-hour meeting transcript = 10,000–20,000 tokens 1M tokens = 750,000 words = 1,500 pages
Do this
- Start with Sonnet for everything; switch only with a specific reason
- Drop to Haiku for repetitive tasks — roughly 3x cheaper, just as accurate
- Reach for Gemini only when you need video or audio
- Check your model picker regularly — Anthropic releases new models periodically
Avoid this
- Using Opus for simple questions — 5x the cost, no noticeable quality gain
- Pushing a very long document through Haiku — its smaller context window may truncate it
- Assuming the heaviest model is always best
- Running everything through one model out of habit
For a very long document (over roughly 30 pages), use Sonnet to avoid truncating your content.
Verification
Related Resources
- claude.ai — Open the model picker to see the models available on your plan
- docs.claude.com/docs — Official documentation
- support.claude.com — Help center and account support
Module 1 complete
You now understand how AI models work, how to prompt them well, and how to pick the right one for any task. Open claude.ai, set Sonnet as your default, and run your next real task through the decision flow.