Claude
Anthropic's family of AI assistants: Transformer-based large language models trained with RLHF and Constitutional AI to be helpful, honest and harmless.
Anthropic, 2023
A chat assistant that has read a huge amount of text and then been coached, partly using a written list of principles, to be helpful and to avoid causing harm.
Where it fits
+All relationships as a list
Is a kind of / part of
Siblings
Uses / built on
How it works
Claude is a family of assistant models made by Anthropic, first announced on 14 March 2023. Model versions and capabilities change quickly, so this page sticks to the durable facts.
Under the hood, Claude is a large language model: a Transformer pretrained to predict the next token on large amounts of text, then post-trained to act as an assistant.
What sets its training apart is Constitutional AI. Instead of relying only on human ratings, the model is given a written set of principles. It critiques and revises its own drafts against those principles, and AI-generated preference judgments based on them feed a reinforcement learning stage, alongside human feedback. See RLHF for how that stage works.
Newer generations of Claude accept images as well as text, so they can read charts, screenshots and documents.
Everyday use cases
- Case 01
How you've probably met it
Through the Claude apps, in coding tools, or inside other products that use Anthropic's API for writing, analysis and customer support.
Strengths, weaknesses and when to use it
Strengths
- Strong at long documents, writing and coding.
- Trained with an explicit set of written principles (a "constitution").
Weaknesses
- Like all LLMs, it can make mistakes and state wrong things confidently, so important facts should be checked.
When to use itFor conversational help with writing, analysis, coding and working through long documents.
Go deeper
Sources
- [1]Anthropic, Introducing Claude (March 2023)
- [2]Bai et al., Constitutional AI: Harmlessness from AI Feedback (2022)
- [3]Anthropic, Claude
Facts checked as of 2026-10-11