A small model that does real work
We have added Claude Haiku 5.5 to Ayeto. It is the newest model of Anthropic's Claude 5.5 family, and Anthropic calls it its cheapest, fastest and most capable small model to date. On average it costs around 75% less to run than Haiku 4.5.
It is not meant to replace your strongest model. It is built for work that happens thousands of times a day and should be quick and inexpensive.
What is Claude Haiku 5.5?
Claude Haiku 5.5 is designed for high-volume, cost-sensitive tasks. Anthropic points to summaries, compaction, database queries and classification, as well as speed-sensitive work such as live customer support and browser use. It also pairs well with Opus 5.5 and Sonnet 5.5 as a subagent on coding work.
It is the first Haiku with an adjustable reasoning effort, from low to max, so you can decide whether a task should favour cost or intelligence. Reasoning cannot be switched off completely, and the default is medium.
How it compares
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (Elo) | 1620 | 735 | 1437 | 1840 |
| OSWorld 2.1 (offline subset) | 72.4% | 15.7% | 48.9% | 83.9% |
| Humanity's Last Exam (no tools) | 45.9% | 10.2% | — | 56.9% |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 16.4% | 70.6% |
| FrontierCode 1.1 (Main) | 46.4% | — | 42.4% | 52.1% |
Source: Anthropic.
Against Haiku 4.5 and GPT-6 Luna
The jump over Haiku 4.5 is large, and Haiku 5.5 also leads GPT-6 Luna in every benchmark where Luna was measured. In early testing, Asana reported over 30% lower latency for task completions, and Box measured an 11-point quality gain over Haiku 4.5 at about half the latency.
Where bigger models still win
On complex agentic coding such as Terminal-Bench 4.0, Sonnet 5.5 and Opus 5.5 remain clearly stronger. Haiku 5.5 is best suited to narrower, well-defined tasks that were previously too costly to run at scale, such as compaction, summarization or subagent work.
When to choose Haiku 5.5 in Ayeto
- Scheduled tasks and automations that run often and must stay cheap.
- Summaries, data extraction and classification across large volumes of documents or e-mails.
- Quick answers and live conversations, where response time decides the experience.
- Subagents that handle small lookups while a larger model leads the work.
For long coding tasks, demanding research or documents where precision matters most, stay with Sonnet 5.5 or Opus 5.5. In Ayeto, you can switch models in the assistant settings without changing the rest of its setup.
How its cost compares
Per million tokens, Haiku 5.5 costs $0.10 for input and $0.50 for output on prompts up to 100,000 tokens, against $1 and $5 for Haiku 4.5. Above 100,000 tokens the price rises to $0.50 and $2.50, so it offers the best value when the context is small. GPT-6 Luna costs the same up to 100,000 tokens and less beyond that, so for very long prompts it is worth comparing both.
Haiku 5.5 also uses a newer tokenizer, so the same text takes slightly more tokens than before (about 1.25× on a long prompt in one independent test). Anthropic's figure of around 75% already takes this into account, which is why the real saving is somewhat smaller than the gap in the price list alone.
Ayeto converts usage into credits. The most reliable way to decide is to run the same real task through Haiku 5.5, GPT-6 Luna and the model you use today in Ayeto Model Comparison and compare quality, processing time and total credit usage.
Try Claude Haiku 5.5
Claude Haiku 5.5 is now available in Ayeto. Select it in your assistant settings or directly in the chat, and hand it the work you would otherwise send to a more expensive model.