Claude Code · Series · Part 4 of 8

The Claude API, every model so far, and the official links

Parts 1 to 3 were about using Claude Code well. This part is the reference shelf at the end of the beginner level (Parts 5 to 8 go further): how Claude Code differs from the Claude API, a first API call you can run in five minutes, every Claude model published so far with its API ID and status, and the official pages I rely on, all in one place. Models and pages change often, so every table here is a dated snapshot and the linked docs are the source of truth.

1. Claude apps, Claude Code or the API?

The same models are reachable through different doors. Which one you want depends on what you are trying to do:

2. The Claude API in five minutes

If you have read my Android post on online models, this will look familiar: the API takes the whole conversation as JSON and returns the next message. It does not remember anything between calls, so you re-send the history every time (the same idea as the context window in Part 2).

1. Create a Console account at platform.claude.com. This is separate from a claude.ai chat plan and is billed by usage.
2. Create an API key in the Console (see Get an API key) and store it somewhere safe. Treat it like a password.
3. Put the key in an environment variable so it never appears in your code:

   export ANTHROPIC_API_KEY="your-api-key-here"

4. Make a call. Here is the plain HTTP version from the official quickstart:

   curl https://api.anthropic.com/v1/messages \
     -H "content-type: application/json" \
     -H "x-api-key: $ANTHROPIC_API_KEY" \
     -H "anthropic-version: 2023-06-01" \
     -d '{
       "model": "claude-opus-5-5",
       "max_tokens": 1000,
       "messages": [
         { "role": "user", "content": "Explain what a token is in one sentence." }
       ]
     }'

The reply is JSON. The text is in content, stop_reason says why it stopped, and usage reports the input_tokens and output_tokens you were charged for. The same call with the official Python library is shorter:

import anthropic

client = anthropic.Anthropic()   # reads ANTHROPIC_API_KEY

message = client.messages.create(
    model="claude-opus-5-5",
    max_tokens=1000,
    messages=[
        {"role": "user", "content": "Explain what a token is in one sentence."}
    ],
)

for block in message.content:
    if block.type == "text":
        print(block.text)

Official libraries exist for Python, TypeScript, C#, Go, Java, PHP and Ruby, plus a command-line tool, listed on the client SDKs page. The next page to read is Working with the Messages API: multi-turn conversations, system prompts and stop reasons.

3. Every Claude model published so far

This is the full lineup as the official docs list it on 14 October 2026. Anthropic retires old models regularly, so treat it as a snapshot and check the models overview and the model deprecations page for the live status. “MTok” means a million tokens.

Current lineup

All four accept text and image input, have a 1M-token context window and up to 128K tokens of output on the standard API, and are listed as not retiring sooner than September to October 2027. Prices are Claude API base prices and are separate from a Claude subscription. Two discounts from the pricing notes: batch requests are 50% off, and prompt-cache reads cost a fraction of the base input price (10% in general, and 2.5% to 5% on some of these models), which is why a stable CLAUDE.md is cheap to re-read (Part 3). See the pricing page for cache writes, long-context and per-platform rates.

Other models still active

The overview page calls Fable 5, Opus 5 and 4.x, Sonnet 5 and 4.6 and Haiku 4.5 “legacy”: still available, but no longer receiving updates. The Mythos models are listed as active in the deprecations table, but I could not find a plain description of them, so see the docs before relying on them.

Deprecated and retired

IDs, aliases and other platforms

• Every Claude model ID is a pinned snapshot. From the 4.6 generation on, IDs have no date suffix (for example claude-sonnet-5-5) and are still a fixed snapshot. Older models have dated IDs such as claude-haiku-4-5-20251001.
• Claude Code aliases like sonnet or opus are shortcuts to the latest version (Part 1, section 7). Use a full ID when you need repeatable behaviour.
• Other platforms: Google Cloud, Microsoft Foundry and Claude Platform on AWS use the same IDs as the Claude API. Amazon Bedrock prefixes them, for example anthropic.claude-opus-5-5, and Bedrock and Google Cloud set their own retirement dates.
• Ask the API itself. The Models API returns the models your key can use, with their token limits and capabilities, so you do not have to hard-code a list.

4. Choosing between models

The official guidance, put plainly:

Two sources, two slightly different defaults: the API guide starts you at Opus 5.5, while the Claude Code cost guide suggests Sonnet for everyday coding and Opus for complex architecture. Both are consistent with one rule: start with the model your plan defaults to, move down to save time and money when the task is easy, and move up when you are stuck. Test on your own tasks, because a cheaper model that passes your checks is the better choice. The docs have a choosing a model guide, and the migration guide helps when you move to a newer generation.

5. Official links

All of these were reachable when I wrote this post. Anthropic reorganises its docs from time to time (the Claude Code docs, for example, moved to code.claude.com), so if a link breaks, search the docs index at code.claude.com/docs/llms.txt.

Claude Code: getting started

Claude Code: memory, settings and control

Claude API and models

Plans, learning and help

6. Beginner-level wrap-up and glossary

1. Part 1: set up Claude Code safely, learn the modes, and pick a model.
2. Part 2: give Claude lasting context with CLAUDE.md, rules and skills.
3. Part 3: spend fewer tokens, hand work off between sessions and onboard faster.
4. Part 4: use the API when you want Claude inside your own product, and keep the model list and links handy.
5. Next (advanced): Part 5 covers day-to-day prompting, git and troubleshooting.

The model forgets by design and the docs change often. Writing things down, and checking the official pages when something looks off, are the two habits that carry through all four parts.