Names, versions and owners change fast. Checked October 2026.
In plain words
Claude is two things with one name. It is a family of large language models built by the American company Anthropic. It is also the app, on the web, desktop and phone, where you chat with those models.
The models come in sizes. Smaller ones answer quickly and cheaply. Larger ones are slower but handle long, difficult work such as writing software or analysing a stack of documents. In the app you can usually pick which one to use.
Each tier has a name, such as Haiku for the quickest models and Opus for long, demanding work, and each gets new version numbers over time; the box above lists the current ones. The first Claude came out in March 2023. Like ChatGPT and Gemini, it is a chatbot on the surface and a transformer-based model underneath.
Everyday example
Choosing a cook for the occasion
For a quick cup of chai you call the person who is fastest in the kitchen. For a wedding feast you call the most experienced cook, who takes longer but gets every dish right. Claude's tiers work the same way: a small, quick model for short tasks, a middle one for everyday work, and the largest ones for big, careful jobs.
Where the example breaks: Cooks have skills of their own. Claude's tiers are not separate personalities: they trade speed and cost against capability, so even the small one can surprise you and the big one can still make mistakes.
Latestas of October 2026
- Current models
- Claude Haiku 5.5 (7 October 2026), Sonnet 5.5 (28 September 2026), Opus 5.5 (22 September 2026) and Fable 5.1
- Restricted model
- Claude Mythos 5, the same model as Fable 5 with some safeguards lifted, offered only to selected cybersecurity partners since June 2026
- Maker
- Anthropic, which filed on 26 October 2026 to list its shares on the Nasdaq stock market
These details change often. The rest of this page explains what it is, which stays true.
How it actually works
Claude is built the same broad way as other modern chatbots, with a few methods Anthropic has written about publicly.
Pre-training
Each model first learns to predict the next token across a very large collection of text and code. That gives it general knowledge and fluent language. This is pre-training.
A written constitution
Anthropic shapes Claude's behaviour partly with a written set of principles. The model is trained to critique and revise answers against those principles, alongside human feedback. Anthropic calls this Constitutional AI, a form of alignment work.
Thinking and tools
Newer Claude models can work through a problem step by step before answering, and can use tools: searching the web, running code, or operating a computer screen. See tool use and computer use.
- Made by
- Anthropic, San Francisco
- First released
- 14 March 2023
- Model tiers
- Several sizes, from small and fast to large and deliberate, each with its own version number
- Training approach
- Pre-training, then human feedback and Constitutional AI
Anthropic also offers some models only to selected organisations, such as cybersecurity partners, rather than to the general public. The box above has the current details.
Questions people ask
Is Claude the same as ChatGPT?
Who makes Claude?
Anthropic, an American AI company founded in 2021 by former OpenAI staff, including siblings Dario and Daniela Amodei. It is organised as a public benefit corporation and is based in San Francisco.
What do Haiku, Sonnet, Opus and Fable mean?
They are the names of Claude's tiers, from fastest to most capable. Haiku suits quick, high-volume tasks, Sonnet balances speed and ability, Opus handles long, complex work, and Fable, added in 2026, sits above it. The number after the name is the version.
Can Claude use a computer or browse the web?
It can when the app or developer turns those tools on. Anthropic released a "computer use" ability in October 2024 that lets Claude look at a screen, move a cursor and type. Claude can also search the web and run code to check its work.
Can Claude make mistakes?
Yes. Like every large language model, it can state wrong things confidently, which is called hallucination. Check important facts, and treat its answers as a helpful draft rather than the final word.