Claude's Model Lineup
Anthropic doesn't offer one Claude — they offer a family of models at different points on the capability-speed-cost spectrum. Choosing the wrong model for a task is one of the most common ways people underuse Claude. This module gives you the exact decision framework you need to always use the right tool.
The Claude family explained
As of 2025, Anthropic's Claude family consists of three tiers: Haiku, Sonnet, and Opus. Each name comes from a type of poetry — Haiku being short and precise, Sonnet being structured and expressive, Opus being a major work of depth and complexity. The naming is deliberate.
Each tier has a numbered version (Claude 3, Claude 3.5, Claude 4) indicating the generation. A Claude 4 Haiku is faster and cheaper than a Claude 3 Sonnet, but may be less capable on complex reasoning tasks. Understanding both the tier and the generation is essential for making good model choices.
Haiku is Claude's speed-optimized tier. It's designed for tasks where response time matters more than maximum depth — high-volume applications, real-time interactions, quick lookups, and lightweight tasks that don't require heavy reasoning. Don't mistake fast for dumb — Haiku is still remarkably capable. It handles most everyday tasks well. What it doesn't do is match Sonnet or Opus on complex multi-step reasoning, nuanced writing, or tasks requiring deep context integration.
Sonnet is the sweet spot — and for most users most of the time, it's the right choice. It balances capability, speed, and cost in a way that makes it the default workhorse for serious work. Claude Sonnet 4 and 4.5 are among the most capable models available from any company for practical everyday tasks. When people say "I use Claude," they're almost always using Sonnet. It handles complex reasoning, long documents, nuanced writing, and multi-step tasks with high reliability.
Opus is Anthropic's most capable model — their frontier, the one they use to demonstrate what's possible. It excels at tasks requiring the deepest reasoning, the most nuanced judgment, and the most sophisticated synthesis. It's slower and significantly more expensive than Sonnet, which means it's not the right default. But for tasks where you genuinely need maximum capability — complex research, difficult code, high-stakes writing, deep strategic analysis — Opus is worth the premium.
Understanding model versions — the numbers matter
Beyond Haiku/Sonnet/Opus, Claude models have version numbers. Claude 3 models launched in early 2024. Claude 3.5 models launched mid-2024 with significant improvements. Claude 3.7 followed with reasoning enhancements. Claude 4 models launched in 2025 and represent a substantial generational leap.
A key insight: newer Haiku models often outperform older Sonnet or Opus models on many tasks. Claude Haiku 4.5 may outperform Claude Sonnet 3 on standard benchmarks despite being the lower tier in the newer generation. This matters when choosing models through the API or comparing options.
| Model | Tier | Best version string | Primary strength |
|---|---|---|---|
| Claude Haiku 4.5 | Speed | claude-haiku-4-5-20251001 | Fastest, cheapest, high volume |
| Claude Sonnet 4.5 | Balanced | claude-sonnet-4-5 | Best everyday workhorse |
| Claude Sonnet 4 | Balanced | claude-sonnet-4 | Strong reasoning, coding |
| Claude Opus 4.5 | Power | claude-opus-4-5 | Maximum capability |
| Claude Opus 4 | Power | claude-opus-4 | Deep analysis, frontier tasks |
The model selection decision framework
Here's how to think about model selection in practice. Run through these questions in order:
Extended thinking — a special capability
Some Claude models support extended thinking — a mode where Claude works through a problem step by step internally before giving its answer. This is different from regular responses. When extended thinking is enabled, Claude essentially shows its work, reasoning through the problem explicitly before arriving at a conclusion.
Extended thinking is most valuable for:
Hard math and logic problems where the reasoning chain matters as much as the answer.
Complex code debugging where understanding the thought process helps you trust and extend the solution.
Multi-step decisions where you want to see the trade-off analysis, not just the recommendation.
Ambiguous problems where different interpretations lead to different answers and you want to see Claude navigate that explicitly.
In Claude.ai, you can access extended thinking by starting your prompt with phrases like "think through this carefully step by step" or by enabling it in model settings where available. The response will show Claude's reasoning process before its final answer. This is slower but often significantly more accurate on complex tasks.
How to access different models
Claude.ai (web and mobile)
When using claude.ai directly, you select the model from a dropdown in the conversation interface. Claude Pro subscribers have access to Sonnet and Opus. The free tier typically runs on Haiku or an older Sonnet version. If you're on Pro and doing serious work, always verify you're using your intended model before starting.
Claude API
When building with the API, you specify the model using its exact string identifier. Using a specific version string like claude-sonnet-4-5 locks you to that exact model. Using claude-sonnet-4-5-latest automatically uses the newest version of that model tier as Anthropic updates it. For production applications, specific version strings are safer. For personal projects, latest is usually fine.
For most work sessions, start with Sonnet. If you hit a wall — a reasoning problem that feels underserved, a piece of writing that isn't quite landing, a coding challenge Sonnet isn't cracking — switch to Opus for that specific problem. Then switch back. This gives you the speed and cost benefits of Sonnet for routine work while reserving Opus's power for where it actually matters.