Model Families and Current Models¶

Level: Beginner · Time: ~35 min · Exam relevance: CCAO-F D3 (3.2) · CCAR-P D2 (2.1)

In plain English¶

Anthropic releases Claude models in families (also called lines). Each line is a size and speed class, and each new generation of a line replaces the previous one:

Haiku: the smallest and fastest. Good for simple, high-volume work.
Sonnet: the balance of speed and intelligence. A strong everyday choice.
Opus: highly capable, for complex, long-running knowledge work and coding.
Fable: Anthropic's frontier line, for the most demanding reasoning and long-horizon agentic work, at a higher price.

An analogy: a company's travel options. Haiku is the bicycle courier: quick, cheap, perfect for short routine trips. Sonnet is the company car: handles most journeys well. Opus is the experienced driver for long, complicated trips. Fable is the specialist charter for the hardest missions. You do not book a charter to post a letter.

The names stay the same across generations; the version number tells you the generation. "Claude Sonnet 5.5" is the 5.5 generation of the Sonnet line.

Why it matters¶

CCAO-F objective 3.2 asks you to differentiate the Claude model families, and CCAR-P objective 2.1 to select models based on trade-offs. Exams test the pattern (smaller is faster and cheaper; larger is more capable on complex work), not memorised prices. But in practice you also need to know where to find current facts, because model line-ups change several times a year.

How it works¶

Current models (as of 2026-10-24)¶

These figures come from Anthropic's models overview and pricing pages, checked on 2026-10-24. Prices are for the Claude API, per million tokens (MTok).

	Claude Haiku 5.5	Claude Sonnet 5.5	Claude Opus 5.5	Claude Fable 5.1
Positioning	High-volume, latency-sensitive tasks such as classification, extraction and routing	Best combination of speed and intelligence	Long-running agentic coding and knowledge work; Anthropic's suggested starting point for most workloads	Demanding reasoning and long-horizon agentic work
Relative latency	Fastest	Fast	Moderate	Slower
API model ID	`claude-haiku-5-5`	`claude-sonnet-5-5`	`claude-opus-5-5`	`claude-fable-5-1`
Input price	From $0.10 / MTok	$2 / MTok	$4 / MTok	$10 / MTok
Output price	From $0.50 / MTok	$10 / MTok	$20 / MTok	$50 / MTok
Context window	1M tokens	1M tokens	1M tokens	1M tokens
Maximum output	128K tokens	128K tokens	128K tokens	128K tokens
Reliable knowledge cutoff	June 2026	June 2026	June 2026	June 2026

Notes on the table:

Claude Haiku 5.5's listed price applies to prompts up to 100,000 tokens; longer prompts cost more ($0.50 input and $2.50 output per MTok as of 2026-10-24).
The Message Batches API is 50% cheaper for work that can wait. Reading from the prompt cache costs a fraction of the base input price (Prompt caching).
All current models accept text and image input, produce text output, are multilingual, and support tool use.
Thinking is adaptive: the model decides how much to reason, steered by an effort setting (low, medium, high, xhigh, max). Effort is a second dial alongside model choice, covered in Model selection.
Prices on Amazon Bedrock and Google Cloud Vertex AI are set by those providers; Microsoft Foundry is billed at Anthropic's standard API rates.

Older models remain available for a while as legacy models (as of 2026-10-24 these include Claude Fable 5, Claude Opus 5, Claude Opus 4.8, 4.7, 4.6 and 4.5, Claude Sonnet 5 and 4.6, and Claude Haiku 4.5). Each model has a published retirement commitment on the model deprecations page. Anthropic also runs some models under restricted access programmes; those are outside the scope of this repository.

Model IDs¶

In the API you select a model by its ID, such as claude-opus-5-5. From the 4.6 generation onwards, IDs have no date suffix, and each ID is a pinned snapshot: the model behind it does not silently change. On Amazon Bedrock the same models carry an anthropic. prefix (for example anthropic.claude-opus-5-5).

In the Claude apps you choose from a model menu by name, and which models appear depends on your plan (for example, Opus models are not included on the Free plan as of 2026-10-24).

How the families relate¶

quadrantChart
    title Relative positioning of current Claude lines (illustrative)
    x-axis Lower cost per token --> Higher cost per token
    y-axis Simpler tasks --> Most demanding tasks
    quadrant-1 Frontier
    quadrant-2 Best value for hard work
    quadrant-3 High-volume workhorse
    quadrant-4 Rarely the right choice
    Haiku: [0.08, 0.2]
    Sonnet: [0.35, 0.55]
    Opus: [0.55, 0.78]
    Fable: [0.9, 0.93]

The chart is illustrative, not a benchmark. Real performance depends on the task, the prompt and the effort setting. The only reliable way to know which model is best for your task is to test, which Model selection explains.

Generations improve every line¶

A newer Haiku can outperform an older Sonnet on some tasks; a newer Opus at lower effort may match an older model at higher effort. When a new generation arrives:

Read the release notes and migration guide for behavioural changes.
Re-run your evaluation set (Module 14).
Re-check costs, because prices and tokenisers can change between generations.

Model upgrades and change management covers this for production systems.

Finding current facts yourself¶

You need	Go to
Current models, IDs, context windows, cutoffs	Models overview
Prices, batch and caching discounts	Pricing
Retirement dates	Model deprecations
Guidance on choosing	Choosing a model
What's in the apps on your plan	claude.com/pricing and the model menu in the app
Programmatic list of models and limits	The Models API (`client.models.list()`), which reports each model's context window and capabilities

Example¶

Listing the available models from code is a good way to avoid hardcoding stale assumptions. With the official Python SDK:

import anthropic

client = anthropic.Anthropic()  # reads ANTHROPIC_API_KEY from the environment
for model in client.models.list():
    # max_input_tokens is the context window; max_tokens is the output cap
    print(model.id, model.display_name, model.max_input_tokens, model.max_tokens)

Cost: listing models needs an API key but does not run the model.

Enterprise use case¶

Acme Logistics runs three Claude workloads:

Workload	Volume	Needs	Model chosen
Tagging inbound shipment emails by type	40,000 per day	Fast, cheap, simple categories	Claude Haiku 5.5
Drafting customer replies to delivery complaints	1,500 per day	Good tone and accuracy, moderate speed	Claude Sonnet 5.5
Monthly analysis of carrier contract performance across hundreds of pages	A few per month	Deep reasoning over long documents	Claude Opus 5.5 at high effort

The architect documents each choice with the evaluation results that justified it, and schedules a re-evaluation when new generations are released.

Common mistakes¶

Memorising prices for the exam or for planning. Learn the pattern; look up the numbers.
Assuming the biggest model is always best. For simple, high-volume tasks, a smaller model is often equally accurate and far cheaper and faster.
Hardcoding old model IDs and forgetting them. Legacy models are eventually retired; track deprecations.
Comparing models on one example. Compare on a representative set of tasks.
Ignoring effort. On current models, adjusting effort can change cost and quality as much as switching model.

Security considerations¶

Model choice can affect safety behaviour: different models and generations may respond differently to borderline requests and to prompt-injection attempts. Include security test cases in evaluations when switching models (Safety and red-team evals).
Restrict which models teams can use where cost or data-handling commitments differ (for example, some models have specific data-retention requirements). Check the model's page before using it with sensitive data.

Troubleshooting¶

Symptom	Likely cause	What to do
"Model not found" error in the API	Wrong ID, retired model, or model not enabled on your platform	Check the models overview and your platform's model access settings
A model is missing from the app menu	Not included in your plan, or disabled by an admin	Check your plan and organisation settings
Costs jumped after switching model	Higher per-token price or different tokeniser	Re-measure tokens and cost per task

Check your understanding¶

Name the four current Claude lines from fastest to most capable.
Which line does Anthropic suggest as the starting point for most workloads, as of 2026-10-24?
Where would you check the current context window and price for a model?
Why can a newer, smaller model sometimes beat an older, larger one?
Besides choosing a model, what other setting changes the cost-quality balance on current models?

Answers

Haiku, Sonnet, Opus, Fable.
Opus (Claude Opus 5.5).
The models overview and pricing pages in Anthropic's documentation (or the Models API for context windows).
Because each generation improves every line; a newer generation's training and techniques can outweigh an older model's size advantage on some tasks.
The effort level, which steers how much the model thinks.