Claude Explained: Haiku, Sonnet, Opus and Fable (and Why the Names Look Like That)

Four poetic names, one capability ladder. What each Claude model is for, what the naming tells you, how the subscription differs from the API, and what Claude Code adds.

Anthropic names its models like a literature class: Haiku, Sonnet, Opus, Fable. It sounds whimsical, but the names carry real information. Once you decode them, choosing a Claude model takes five seconds.

Snapshot

Checked on 30 September 2026 against Anthropic's own documentation. Model lineups move quickly, so confirm on the models overview.

The naming, decoded

Anthropic introduced Haiku, Sonnet and Opus in 2024 and ordered them by size and capability, like written works that get longer:

Haiku: a short poem. The smallest, fastest, cheapest model.
Sonnet: a longer, structured poem. The balanced workhorse.
Opus: a large work. The most capable of the classic trio, but slower and pricier.

A version number follows (Sonnet 5.5, Opus 5.5, Haiku 4.5). Different tiers can have different version numbers at the same moment, because the small model is not always re-trained at the same time as the large one. So "Haiku 4.5" sitting beside "Opus 5.5" is normal, not a typo.

In 2026 a new name joined the ladder: Fable. According to Anthropic's announcement as reported in trade coverage, Fable sits above Opus and is a model from a more capable "Mythos" class that has been made safe for general release, with safeguards that route queries on some sensitive topics to a different model. Two layers of naming then appear: the class says which capability generation a model belongs to, and the tier (Haiku, Sonnet, Opus) says where it sits between speed and depth. The naming is still new, so read Anthropic's own pages for the current positioning.

The current lineup

From Anthropic's models overview:

Model	Anthropic's description	API price per million tokens (input / output)	Context window
Claude Fable 5.1	demanding reasoning and long-horizon agentic work	$10 / $50	1M tokens
Claude Opus 5.5	long-running agentic coding and knowledge work	$4 / $20	1M tokens
Claude Sonnet 5.5	the best combination of speed and intelligence	$2 / $10	1M tokens
Claude Haiku 4.5	the fastest model with near-frontier intelligence	$1 / $5	200K tokens

The API IDs are simple: claude-fable-5-1, claude-opus-5-5, claude-sonnet-5-5 and claude-haiku-4-5. Older versions stay available as "legacy" models for a while, which is useful when you need to reproduce old results.

A few details worth knowing:

Thinking and effort. The larger models use adaptive thinking: the model decides how long to reason before answering, steered by an effort setting. Higher effort means better answers on hard problems, more tokens and more latency.
Context window. One million tokens is roughly 555,000 words. That is a whole codebase or several books in one prompt.
Discounts. Batch requests cost 50% less. Re-reading a cached prompt costs a fraction of the normal input price.

Which one should you use?

If you want...	Reach for
Quick classification, extraction, routing, high-volume chat	Haiku
An everyday assistant, coding help, drafting and analysis	Sonnet
A hard multi-step job: a large refactor, deep research, long agent runs	Opus
The toughest reasoning when Opus at high effort still falls short	Fable

A sensible habit: start at Sonnet, move up only when the answers are not good enough, and move down to Haiku when you are running something thousands of times and quality is already fine. Anthropic's own guidance in the overview suggests starting with Opus for most workloads and using Fable when evaluations on Opus still fall short.

What you can do with Claude

Chat and write: long documents, careful editing, explanations, brainstorming.
Read at scale: upload contracts, papers or data and ask questions.
Code: generate, explain, review and debug.
Build apps: the API supports tool use, vision, structured output and long-running agents.
Work as an agent: with Claude Code, Claude reads your repository, edits files, runs commands and tests, and commits changes.

Subscription versus API

These are two separate doors into the same models.

The subscription (claude.ai and the apps). You pay a flat monthly fee and get a usage allowance. On the pricing page the individual plans are:

Free: Sonnet and Haiku in chat.
Pro ($20 a month, or $17 a month billed annually): adds Claude Code and access to Opus and Fable, with limited Fable usage.
Max (from $100 a month): the same as Pro with 5 times or 20 times the usage, for people who use Claude all day.
Team and Enterprise: seats with admin controls, single sign-on and central billing. Enterprise adds usage billed at API rates.

Subscription limits reset on a schedule (the pricing page describes usage per five-hour session), so you never get a surprise bill, but you can run out mid-afternoon.

The API (the Claude Console). You create an account on the developer platform, get an API key, and pay per token at the rates above. There is no monthly cap other than the spending limit you set. Use it when you are building your own software, or when you want exact control over the model, the prompt and the cost. Full rates, including batch, caching and long-context pricing, are in the API pricing docs.

The rule of thumb: people use the subscription, programs use the API. A subscription does not come with API credits, and an API key does not unlock the consumer app.

Claude Code: Claude with hands

Claude Code is Anthropic's agentic coding tool. You start it in a project folder:

cd your-project
claude

Then you talk to it in plain language ("write tests for the auth module, run them, and fix any failures") and it plans, edits files across the project, runs commands and checks the result.

Where it lives:

Terminal CLI, the full-featured version.
VS Code and JetBrains extensions, with inline diffs.
Desktop app, for several sessions side by side and scheduled tasks.
Web and mobile, for long tasks that run in the cloud.

What makes it more than a chat box:

Project memory. A CLAUDE.md file in your repository holds your standards and preferences, and Claude reads it at the start of each session.
Permissions and plan mode. It asks before risky actions, and in plan mode it proposes an approach before touching any file.
Subagents and background agents, which split work into parallel pieces.
Skills and hooks, which package repeatable workflows and run your own commands before or after actions.
MCP, the open Model Context Protocol, which connects it to tools like Jira, Slack or your database.
Headless mode for scripts and CI: claude -p "translate new strings and raise a PR".

Claude Code signs in with a Claude subscription or a Console account. If an ANTHROPIC_API_KEY is set, it uses that key and bills API rates instead. The terminal tool, VS Code and JetBrains can also use third-party cloud providers.

Why this matters

Claude is a good example of the whole map: one family of models (Haiku to Fable), two ways in (subscription or API), and one agent (Claude Code) that runs on several surfaces. The next articles show the same pattern for OpenAI and Google.