Anthropic released Claude Haiku 5.5 on 7 October 2026. It is the company's small, fastest model, and it comes with a steep price cut: $0.10 per million input tokens, down from $1 for Haiku 4.5. In its official announcement, Anthropic calls it "the cheapest, fastest, and most capable small model" it has released. Here is what it is, how its price compares with Haiku 4.5 and Sonnet 5.5, when to use it and how to try it today.
What is Claude Haiku 5.5?
Haiku is Anthropic's lightweight model line, built for high-volume work where latency and cost matter most. According to the official documentation, these are the key specs:
- Model ID: claude-haiku-5-5 on the Claude API, Google Cloud and Microsoft Foundry, and anthropic.claude-haiku-5-5 on Amazon Bedrock.
- 1 million-token context window and up to 128,000 output tokens (300,000 on the Batch API with a beta header).
- Adaptive thinking with effort control. It is the first Haiku model with an adjustable effort setting, from low to max. The default is medium.
- Text and image input, text output, plus computer use and browser use.
- Reliable knowledge up to June 2026. Anthropic commits to not retiring it before 7 October 2027.
It is available on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.
How much does Claude Haiku 5.5 cost?
Haiku 5.5 has two price tiers based on prompt length: one for requests up to 100,000 tokens and a higher one above that. Here is how it compares with its predecessor and with Sonnet 5.5, in dollars per million tokens:
| Item | Haiku 5.5 (up to 100K / over 100K) | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|
| Input | 0.10 / 0.50 | 1.00 | 2.00 |
| Output | 0.50 / 2.50 | 5.00 | 10.00 |
| Cache write (5 min) | 0.125 / 0.625 | 1.25 | 2.50 |
| Cache read | 0.01 / 0.05 | 0.10 | 0.10 |
Per token, that is a 90% cut for shorter prompts and 50% for prompts over 100,000 tokens. Anthropic sums it up as "about 75% less on average" than Haiku 4.5. One technical caveat: Haiku 5.5 uses the newer tokenizer introduced with Claude 4.7, which counts the same text as roughly 30% more tokens than Haiku 4.5 did. The real saving is large, but somewhat smaller than the table suggests. The Batch API adds a further 50% discount.
A worked example
Say you classify 10,000 support tickets of about 1,500 tokens each, with a 200-token answer. That is 15 million input tokens and 2 million output tokens. With Haiku 5.5 the bill is about $2.50 ($1.50 input, $1 output). With Sonnet 5.5 it is about $50. Through the Batch API, the Haiku figure halves.
Anthropic also used the launch to halve Sonnet 5.5's cache-read price, from $0.20 to $0.10 per million tokens, which it says cuts the cost of most agentic tasks on that model by around 20%.
Performance: how does it compare with Haiku 4.5 and Sonnet 5.5?
Anthropic's published numbers show a huge jump over Haiku 4.5, while Sonnet 5.5 stays ahead on every test:
| Benchmark | Haiku 5.5 | Haiku 4.5 | Sonnet 5.5 |
|---|---|---|---|
| OSWorld 2.1 (computer use) | 72.4% | 15.7% | 83.9% |
| Humanity's Last Exam (no tools) | 45.9% | 10.2% | 56.9% |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 70.6% |
| GDPval-AA v2.1 (Elo score) | 1620 | 735 | 1840 |
These are vendor-published results, so treat them as a guide and test on your own workloads. The overall picture is clear: Haiku 5.5 gets close to Sonnet on knowledge work and computer use, but trails it by a wide margin on agentic coding in the terminal.
When should you use Haiku 5.5 instead of Sonnet 5.5?
Anthropic itself says Sonnet 5.5 and Opus 5.5 remain the better choice for complex agentic coding. Haiku 5.5 is built for a different job:
- Classification, extraction and routing at scale: tickets, emails, forms.
- Summarization and compaction of long conversations or documents.
- Subagents: narrow tasks that a lead model such as Sonnet or Opus hands off to save time and money.
- High-traffic chatbots and assistants, where latency matters more than the last point of quality.
- Browser or computer automation for simple, repetitive flows.
Pick Sonnet 5.5 when the task needs long reasoning, multi-step coding or a very low error rate. A common pattern is to combine them: Haiku filters and prepares, and Sonnet handles the hard part.
How to try Claude Haiku 5.5
- Create an account in the Claude Console and generate an API key.
- Call the Messages API with claude-haiku-5-5. If you are starting from scratch, our guide to building a chatbot with Python and the Claude API works by changing only the model name.
- Remove temperature, top_p and top_k if you are migrating from Haiku 4.5, because the API returns a 400 error with them. Use the effort parameter to control how much the model thinks.
- Check caching and token counts. The minimum cacheable prompt is 512 tokens, and remember the tokenizer counts roughly 30% more tokens.
If you pay for Claude Max or Team, Anthropic now adds a monthly API credit: $100 on Max 5x, $200 on Max 20x and up to $500 on Team, pooled across users. It works with any model, including Haiku 5.5. To automate workflows without building everything yourself, see Claude Managed Agents. For chat use, Claude's free plan includes Haiku and Sonnet models; we share ideas in tips for using Claude for free.
Frequently asked questions
When was Claude Haiku 5.5 released?
Anthropic released it on 7 October 2026, available from day one on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.
How much does Claude Haiku 5.5 cost per million tokens?
$0.10 input and $0.50 output for prompts up to 100,000 tokens, and $0.50 and $2.50 for longer prompts. The Batch API gives a 50% discount.
Is Claude Haiku 5.5 better than Sonnet 5.5?
No. Sonnet 5.5 scores higher on every published benchmark. Haiku 5.5 is far cheaper and faster, and is designed for narrow, high-volume tasks.
What is Claude Haiku 5.5's context window?
One million tokens, with up to 128,000 output tokens on the synchronous API.
