Skip to content
Technology

Claude

The AI model I rely on for demanding work.

Claude by Anthropic is the AI behind my client projects and internal automations. Precise instruction following, reliable code quality and a clear focus on safety - that is why I build on this model.

Book a free consultation More about me
What is Claude

Claude and Anthropic in brief

Claude is the family of AI models built by Anthropic. Anthropic puts AI safety and reliable behaviour at the centre of its work - the Constitutional AI approach results in models that hallucinate less and tend to ask rather than guess when uncertain.

For practical use, one thing matters above all: Claude follows instructions without drifting off creatively. In software development this predictability is not a nice-to-have, it is a prerequisite. That very reliability makes Claude a technology business processes can depend on.

Safety first

Anthropic's focus on AI safety and Constitutional AI produces models that hallucinate less and know their limits.

Precise instruction following

Claude carries out instructions accurately instead of deviating on its own - the basis for predictable results.

Reliability

High consistency across many tasks - from a short text to a multi-step agentic workflow.

Why I use Claude

My reasons for Claude

I have worked with Anthropic's models since the Claude 2 era. The decision to use Claude as the primary model for client projects and internal automation is based on concrete day-to-day experience.

Code quality

With SWE-bench scores above 88 percent (Opus 4.8) and high tool-invocation precision, Claude delivers production-ready code changes that rarely need rework.

Reasoning

Deep, multi-step reasoning for complex problems - with effort-level control, thinking depth can be matched to the task.

Agentic workflows

Claude Code enables fully autonomous development workflows - from analysis through implementation to git push. A serious productivity multiplier.

Privacy in mind

Sensitive data can be processed via EU deployments. Privacy is part of the architecture, not bolted on afterwards.

„Predictability is no luxury in software development. Claude delivers it - and that is exactly why I build on it.“

Current models

The Claude model family

Three clearly defined performance tiers - I pick the right model for each task.

Claude Opus 4.8

Flagship

Maximum intelligence for complex tasks.

  • Current flagship (released 28 May 2026)
  • Context window: 1 million tokens
  • SWE-bench Verified: 88.6 %
  • For agentic workflows and deep reasoning

Claude Sonnet 4.6

Standard

The balanced middle ground for production use.

  • Recommended standard model
  • Context window: 1 million tokens
  • Maximum output: 64,000 tokens
  • For around 90 % of all production workflows

Claude Haiku 4.5

Fast

Speed and efficiency at high volume.

  • Fastest model in the family
  • Context window: 200,000 tokens
  • 4-5x faster than Sonnet
  • For real-time classification and triage
Up to date

I stay close to the model landscape

Anyone using AI in production has to keep an eye on change. A recent example: on 12 June 2026, Claude Fable 5 and its sister model Mythos 5 were shut down worldwide by order of the US Commerce Department - the first time a US AI lab had to pull a publicly available frontier model from the market on official instruction. For me that means: model availability deserves deliberate risk consideration, and I plan projects so that a model outage does not become a business outage.

The rest of the Claude model family is unaffected by the order and remains fully available.

GDPR & use in projects

GDPR-compliant use

For European businesses, GDPR is the central test when choosing an AI provider. Claude can be used compliantly - with the right prerequisites.

A data processing agreement (DPA) under Article 28 GDPR is available from the Team plan onwards. Since the direct Anthropic API does not yet offer native EU data residency, for sensitive data I use access via AWS Bedrock EU or Google Cloud Vertex AI EU - complemented by a documented transfer impact assessment.

DPA from the Team plan

Data processing agreement under Article 28 GDPR - the minimum requirement for business use involving personal data.

EU access via Bedrock / Vertex

AWS Bedrock EU (Frankfurt, Ireland) and Google Cloud Vertex AI EU process data within the EU region.

Clean documentation

A transfer impact assessment and clear rules on which data ever enters a prompt.

FAQ

Frequently asked questions about Claude

Answers to the most important questions on model choice, data protection and use.

Which Claude model do you use for my project?
It depends on the task. As a rule of thumb: Haiku 4.5 for high-volume tasks with short response times, Sonnet 4.6 as the standard for around 90 percent of all production workflows, and Opus 4.8 only for demonstrably complex tasks where Sonnet is not enough. I choose the economically and technically appropriate model for each use case.
Can Claude be used in a GDPR-compliant way?
Yes, with the right prerequisites. A data processing agreement (DPA) under Article 28 GDPR is available from the Team plan onwards; free and Pro plans do not include one. Since the direct Anthropic API does not yet offer native EU data residency, for sensitive data I recommend access via AWS Bedrock EU or Google Cloud Vertex AI EU, complemented by a documented transfer impact assessment.
Why Claude and not another AI model?
Claude wins me over with precise instruction following, high code quality and a clear focus on safety. In software development in particular, predictability is decisive - Claude follows instructions without drifting off creatively. For other requirements I also work with OpenAI and Google Gemini as a complement.
Does using Claude incur high costs?
No. With mechanisms such as prompt caching (up to 90 percent savings on cached tokens) and the Batch API (50 percent discount), API costs can be reduced considerably. A typical client project incurs API costs in the single-digit euro range.
What happens if a model is no longer available?
Model availability is not a given - the shutdown of Claude Fable 5 in June 2026 made that very clear. I therefore plan projects with defined fallback models and documented dependencies, so that a model outage does not become a business outage.

Want to use Claude in your business?

In a free initial consultation we look together at where AI creates the most value for you - from model choice through integration to GDPR-compliant implementation.

Book a free consultation