Claude Haiku 5.5 in the EU, processed in European data centres

Claude Haiku 5.5 is a closed model from Anthropic, released on 7 October 2026 as the third model in the Claude 5.5 family. Anthropic describes it as its fastest model to date, built for high-volume, latency-sensitive tasks. It reads text and images and writes text, with a context window of one million tokens and up to 128K tokens per response.

Haiku 5.5 supports adaptive thinking, steered by an effort setting whose default is medium. It uses the newer Claude tokenizer, so the same text counts as roughly 30 percent more tokens than on Claude Haiku 4.5. Anthropic gives its reliable knowledge cutoff as June 2026.

Anthropic positions Haiku 5.5 for quick, repetitive work: classification, routing, extraction, summaries, compacting long conversations and writing database queries. Its speed suits interactive uses such as live customer support, and it works well as a subagent that Claude Opus or Sonnet hands smaller coding tasks to.

It is the smallest model in the Claude 5.5 family, so it is not the right choice for long, complex reasoning, large refactors or tasks where a wrong answer is costly. For those, use Claude Sonnet 5.5 or Claude Opus 5. A common pattern is to let Haiku 5.5 handle the bulk of requests and pass the hard ones to a larger model.

inference.eu speaks the Anthropic Messages format, so the official Anthropic SDKs work once you change the base URL and the key. In Claude Code, set ANTHROPIC_BASE_URL to the inference.eu host and ANTHROPIC_AUTH_TOKEN to your inference.eu key, then use claude-haiku-5.5 wherever you want the small, fast model.

The same model is available through the OpenAI-compatible endpoint, so the OpenAI SDKs, LangChain, LlamaIndex and the Vercel AI SDK work with the model set to claude-haiku-5.5. Anthropic advises leaving temperature, top_p and top_k unset for this model. Ask for JSON in your prompt or use tool definitions when you need structured output.

When you call Claude Haiku 5.5 on inference.eu, your prompts, images and the model's responses are processed in data centres in the European Union. They are used to answer your request and are never used to train any model. That matters for high-volume work, where the inputs are often customer messages, tickets and documents that contain personal data.

We sign one data processing agreement under Article 28 of the GDPR with every customer and provide our list of subprocessors, so your data protection officer can see who handles the data and where. The same agreement covers every model on inference.eu, billed in euros on one invoice.