What is Claude? The LLM, the apps and the API
What is Claude? Anthropic's model family explained: the four current models, every access path, what it costs, and where Claude can run inside the EU.
TL;DR
Claude is Anthropic’s family of large language models, plus the apps and interfaces you reach them through. Four models are current: Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, all with a one million token context window. You can reach them through claude.ai, Claude Code, the Claude API and three hyperscalers. Only the Bedrock and Google Cloud paths process data inside the EU.
What is Claude?
Claude is a family of large language models built by Anthropic. The name covers three different things that get mixed up in conversation: the models themselves, the end-user apps at claude.ai, and the API that software uses to call the models. Someone who says “Claude” and means the chat window means the app. Someone building Claude into a product means the API.
Technically, every current model takes text and images as input and returns text. All of them support tool use, meaning they can trigger functions in your own software, and all of them think adaptively, deciding for themselves how much compute to spend on an answer. Anthropic’s models overview lists the exact identifier, context window and price for each one.
Which Claude models exist?
Four models are current. Prices are per million tokens:
| Model | API ID | Input | Output | Context | Profile |
|---|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 |
10 US dollars | 50 US dollars | 1M tokens | Demanding reasoning, work spanning days |
| Claude Opus 5.5 | claude-opus-5-5 |
4 US dollars | 20 US dollars | 1M tokens | Agentic coding and knowledge work |
| Claude Sonnet 5.5 | claude-sonnet-5-5 |
2 US dollars | 10 US dollars | 1M tokens | Best mix of speed and intelligence |
| Claude Haiku 5.5 | claude-haiku-5-5 |
from 0.10 US dollars | from 0.50 US dollars | 1M tokens | High volume, classification, routing |
All four share the same knowledge cutoff, June 2026, and return at most 128,000 tokens per response. Anthropic recommends Opus 5.5 as the default and Fable 5.1 only where Opus 5.5 at high effort still falls short. Older generations stay available as legacy models, including Opus 5, Opus 4.8, Sonnet 5 and Haiku 4.5.
Two discounts change the bill materially. Batch requests cost half, and prompt cache reads cost 10 percent of the base input price, dropping to 5 percent on Opus 5.5 and Sonnet 5.5 and 2.5 percent on Fable 5.1.
How do you access Claude?
There are seven paths. They do not differ in the models they serve, they differ in billing, contracting party and where your data travels:
- claude.ai as web, desktop and mobile apps. The chat window most people mean.
- Claude Code as a command line tool for development work, included from the Pro plan up.
- The Claude API, with the Messages API for your own software plus Managed Agents for long running jobs.
- The Claude Console at platform.claude.com, the playground and the place to manage keys, workspaces and limits.
- Amazon Bedrock, with AWS as the contracting party and billing on your AWS invoice. Model IDs carry an
anthropic.prefix. - Google Cloud, either as the Google operated offering on Agent Platform or as the Anthropic operated one.
- Microsoft Foundry, billed through Azure Marketplace.
The difference is not cosmetic. Bedrock and Google Cloud are missing features the Claude API has: the Files API, server side tools such as code execution and web search, Agent Skills, the MCP connector and the Batch API. If your build needs one of those, you cannot freely swap the path later.
What does Claude cost?
People at a screen pay per seat, software pays per token. The plans, from the pricing page:
| Plan | Price | Who it is for | Core |
|---|---|---|---|
| Free | 0 US dollars | Individuals | Chat, web search, Artifacts, files |
| Pro | 17 US dollars per month billed annually, 20 monthly | Individuals | Claude Code, Projects, more models |
| Max | from 100 US dollars per month | Heavy users | 5x or 20x Pro usage, 100 to 200 US dollars in API credits |
| Team | 20 US dollars per seat billed annually, 25 monthly | 2 to 150 people | SSO, admin controls, no model training on your content |
| Enterprise | 20 US dollars per seat billed annually plus usage at API rates | 20 seats and up | SCIM, audit logs, Compliance API, custom retention |
An Enterprise invoice has two lines, seats and usage. On developer heavy teams the usage line routinely exceeds the seat line. Our Enterprise versus Bedrock comparison works that arithmetic through.
Does Claude run in the EU?
This is the question most European projects turn on, and the answer depends entirely on which path you take:
| Path | Processing inside the EU | Premium |
|---|---|---|
| Claude API from Anthropic | No. inference_geo accepts only us and global, and workspace geo is fixed to us and cannot be changed after creation |
US only inference costs 1.1x |
| Amazon Bedrock | Yes. Eight EU regions, among them eu-central-1 Frankfurt, eu-west-1 Ireland, eu-west-3 Paris and eu-south-1 Milan, plus the EU inference profile |
Regional endpoints 10 percent above global |
| Google Cloud, Google operated | Yes. The eu multi region endpoint and single regions such as europe-west1 |
Multi region and regional 10 percent above global |
| Claude in Microsoft Foundry | No. Only swedencentral in Europe, and no EU data zone is available for Claude |
not applicable |
Two traps deserve attention. First, AWS warns explicitly against reading the processing location off the geographic prefix of an inference profile ID; if you carry residency obligations, query the destination regions from every source region through the API instead. Second, Google Cloud’s regional endpoints serve Sonnet 4.6 and earlier only. For the current models you are left with the global or the eu endpoint there.
Who is Anthropic?
Anthropic is the company behind Claude and describes itself as a public benefit corporation. The mission it states on its company page is to ensure the world safely makes the transition through transformative AI. It treats safety research as a discipline of its own, from interpretability through alignment.
One honest limit: the founding year, the founders and the headquarters are not on that company page, and this entry carries only what the primary sources checked in this run actually support.
What Claude is not
Four confusions cost project time again and again:
- Not an open weight model. The weights are not downloadable. Running Claude on your own hardware is not an option, unlike Llama or Mistral.
- Not a search engine. Web search is a tool the models can call, not part of the model. Without it, their knowledge stops in June 2026.
- Not a single model. “Claude” with no version in a requirements document is useless. Price, speed and capability differ between Haiku 5.5 and Fable 5.1 by a factor of 100 on input.
- Not the same thing as ChatGPT or Copilot. Different vendor, different contracts, different data paths. The head to head comparison spells out where they diverge.
FAQ
Every figure here comes from the primary sources as of 29 October 2026. Anthropic changes model names, prices and regional coverage several times a year, so check the documentation before you sign anything.
What is Claude AI?
The same thing as Claude. “Claude AI” is the colloquial form that grew out of the claude.ai domain where the chat apps live. Anthropic calls the product Claude, the model family Claude, and the interface the Claude API.
What is Anthropic?
The company that builds and operates Claude, in its own words a public benefit corporation whose mission is to make the transition through transformative AI a safe one. Anthropic sells Claude directly and through Amazon Bedrock, Google Cloud and Microsoft Foundry.
Is Claude free to use?
Yes, on the Free plan at 0 US dollars, with chat, web search, files and Artifacts, though with a limited usage allowance. Claude Code, Projects and the stronger models start at Pro, 17 US dollars per month billed annually.
Which Claude model should I pick?
Anthropic recommends Opus 5.5 as the starting point for most workloads. Sonnet 5.5 costs half as much and handles writing, summarising and everyday tasks. Haiku 5.5 targets high volume work such as classification and routing. Fable 5.1 only pays off once Opus 5.5 at high effort is not enough.
Can Claude run on my own hardware?
No. Anthropic does not publish the model weights, so an on premises deployment is not possible. If you need sovereignty at that level, you end up with open weight families such as Mistral or Llama, and with different capability.
Does Claude keep data inside the EU?
Only through Amazon Bedrock with EU regions, or through the eu endpoint on Google Cloud. Anthropic’s own Claude API offers the us and global inference geos only, with workspace geo fixed to us. Claude in Microsoft Foundry covers only swedencentral in Europe and has no EU data zone for Claude.
Sources
- Anthropic Docs: Models overview (29 October 2026)
- Anthropic Docs: Intro to Claude (29 October 2026)
- Anthropic Docs: Data residency (29 October 2026)
- Anthropic Docs: Claude in Amazon Bedrock (29 October 2026)
- Anthropic Docs: Claude Models on Google Cloud (Google managed) (29 October 2026)
- Claude: Plans and pricing (29 October 2026)
- Anthropic: Company (29 October 2026)
- Microsoft Learn: Foundry Models from partners and community (29 October 2026)
- AWS: Supported Regions and models for inference profiles (29 October 2026)