What is Claude – and what is it used for?

Claude at a glance: vendor, origins, idea

Claude is the family of generative AI models from Anthropic, a US company founded in 2021 with a focus on AI safety. Anthropic was co-founded by former OpenAI researchers, including Dario and Daniela Amodei. The name is commonly said to reference the information theorist Claude Shannon.

Technically, Claude is a large language model: based on training data, it predicts the most likely next piece of text. A defining feature is Anthropic's training approach called Constitutional AI, in which the model is guided by a set of explicit principles to answer in ways that are helpful yet low-risk.

Model family and ways to access it

Claude comes in several model tiers that trade off capability, speed and cost. The tiers carry recurring names; specific version numbers and prices change regularly – the provider's current overview is always authoritative.

• Opus: the most capable tier for complex analysis, coding and demanding reasoning.
• Sonnet: the balanced tier for everyday production work with a good quality-to-speed ratio.
• Haiku: the fast, economical tier for high volumes and latency-sensitive tasks.
• Access: via the claude.ai web interface and apps, via the Anthropic API, and via Amazon Bedrock and Google Cloud Vertex AI.

Claude's strengths

Claude is designed for considered, well-structured answers and precise instruction-following. Hallmarks include a very large context window (several hundred thousand tokens, depending on the model) that allows processing of long documents, plus multimodal input: alongside text, it can analyse images and PDF content.

• Writing and editing: drafts, summaries, translations and tone adjustments across multiple languages.
• Coding: explaining, generating and debugging code; with Claude Code also working agentically in the terminal and in development environments.
• Document analysis: reviewing, comparing and structuring long contracts, reports or datasets.
• Tool and data integration: the open Model Context Protocol (MCP) connects external systems and data sources; features such as Artifacts make it easier to build documents and small apps.

Business use cases

• Customer service: assisting support teams, drafting replies and summarising conversation histories (with human sign-off).
• Knowledge work: research, preparing internal documentation and question-answering over your own content (retrieval-augmented generation).
• Software development: prototyping, code reviews and automating repetitive engineering tasks.
• Marketing and communications: multilingual content, summaries and first drafts – reviewed editorially.
• Agents and automation: multi-step workflows that drive tools and systems in a controlled way via MCP.

Limits and risks

Like all language models, Claude can produce plausible-sounding but incorrect statements ("hallucinations"). Its knowledge ends at a training cutoff; without connected web or data tools, the model has no awareness of current events. Outputs are not deterministic and should always be checked by humans for business-critical decisions.

• Context limit: very long inputs may be truncated or handled imprecisely despite the large window.
• Privacy and confidentiality: enter sensitive data only under clear internal rules and an appropriate contractual framework.
• Cost and lock-in: volume drives cost; vendor dependency should be a deliberate design choice.

Swiss lens: data protection and compliance

For Swiss organisations, using Claude is governed primarily by the revised Data Protection Act (revDSG/nFADP), overseen by the FDPIC (EDÖB). Where EU individuals are involved or systems are used there, the extraterritorial EU AI Act may also apply. Processing on your behalf requires a data processing agreement (DPA), and data localisation should be reviewed – for example via regional deployment through Amazon Bedrock or Google Cloud Vertex AI.

Anthropic states that it does not, by default, use inputs and outputs from its commercial products and API to train its models; for enterprise use, additional commitments such as limited data retention are offered. Always review the provider's current terms and privacy documentation, as these can change.