Strengths
- The cheapest current Claude model: $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, 90% less than Haiku 4.5.
- It ranks first of 22 comparable models on the Artificial Analysis index (36 points) and is fast, at about 241 output tokens per second.
- Good at controlling a computer for a small model: Anthropic reports 72.4% on OSWorld 2.1, versus 48.9% for GPT-6 Luna.
- The first Haiku with adjustable effort, from Low to Max, and with a 1M-token window.
Weaknesses
- Much weaker at complex coding than Sonnet 5.5: 39.2% versus 70.6% on Terminal-Bench 4.0.
- Prompts over 100,000 tokens cost 5 times more ($0.50 / $2.50), so most of the price advantage disappears with long documents.
- It is wordy: it used 300 million output tokens on the Artificial Analysis tests, the most in its class.
- Safety filters are stricter than on Haiku 4.5, and it refuses penetration testing.
- The new tokenizer splits the same text into about 30% more tokens than Haiku 4.5, which eats part of the savings.
Best for
- Classifying texts and requests
- Extracting data from documents
- Routing requests to the right model or person
- Summaries and condensing long conversations
- Helper agent working under Opus or Sonnet
- High volumes of tasks at a low cost
Not ideal for
- Complex coding with AI agents
- Prompts over 100,000 tokens when cost matters
- Hard reasoning tasks
How to use it in Europe
Haiku 5.5 is mainly for developers. You call it through the Anthropic API or through AWS, Google Cloud and Microsoft Azure. The Claude app lists a Haiku model on every plan, but Anthropic does not say whether that is already version 5.5. Slovakia and Czechia are supported countries. The direct API processes data globally or in the US and stores it in the US, with no EU region to choose.
Price compared with similar models
Benchmark results
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index v5.0 (max effort) | 36 (#1 of 22 in its class) | artificialanalysis.ai ↗, October 2026 |
| GDPval-AA v2.1 Elo | 1620 | anthropic.com ↗ (vendor-reported), October 2026 |
| OSWorld 2.1 offline subset | 72.4% | anthropic.com ↗ (vendor-reported), October 2026 |
| Terminal-Bench 4.0 | 39.2% | anthropic.com ↗ (vendor-reported), October 2026 |
| Humanity's Last Exam, with tools | 57.4% | anthropic.com ↗ (vendor-reported), October 2026 |
Benchmarks cover only a narrow slice of ability and vendors often report them at their best settings. Treat them as a rough guide.
More models from this vendor
- Claude Opus 5.5: Anthropic's main model for demanding work, coding and long documents, and the one Anthropic recommends starting with.
- Claude Fable 5.1: Claude's most expensive model, meant for the hardest tasks and long autonomous work when even Opus 5.5 falls short.
- Claude Sonnet 5.5: Claude's balanced model for everyday work and coding. It is faster than Opus 5.5 and costs half as much.
Similar models from other vendors
- DeepSeek V4.1 Flash (DeepSeek): DeepSeek's cheap, fast open-weight model with a 1 million token window. In the app and API, though, your data goes to China.
- Gemini 3.5 Flash-Lite (Google): A very fast, low-cost Google model for large volumes of simple tasks, such as pulling data out of documents.
- Mistral Small 4 (Mistral AI): Mistral's cheap, fast open model under Apache 2.0 for everyday tasks through the API or on your own server.
- GPT-6.1 Terra (OpenAI): OpenAI's fast, cheaper model for working with images, operating a computer and everyday office work.
- Qwen3.8-Flash (Alibaba (Qwen)): A very cheap, fast Qwen model for high volumes of tasks. Unlike Max and Plus, it runs in Frankfurt with processing kept in the EU.
- Grok 4.3 (xAI): An older, cheaper Grok from April 2026 with a 1-million-token window. Good for lots of low-cost API calls.