What is Claude Haiku 5.5? Price, performance and use cases
Anthropic's new small model Haiku 5.5 cuts the price to a tenth while making a big leap in computer use and knowledge work benchmarks. We compiled the numbers, its limits and what it's suited for.
Contents 8
Claude Haiku 5.5 is the fastest and cheapest Claude model, released by Anthropic on October 7, 2026. It is designed for high-volume work such as summarization, classification, data extraction and live customer support. Input starts at $0.10 per million tokens; according to Anthropic, it does an average job for roughly 75 percent less than Haiku 4.5.
In short:
- The API ID is
claude-haiku-5-5; it has a 1 million token context window and a 128K token output limit.- It is 90% cheaper than Haiku 4.5 for requests up to 100K tokens and 50% cheaper for longer ones.
- It is the first Haiku model with an adjustable effort level.
- For complex agentic coding, Anthropic still recommends Sonnet 5.5 and Opus 5.5.
What does Claude Haiku 5.5 bring?
According to Anthropic's announcement, the new model is the cheapest, fastest and most capable small model the company has released so far. In addition to the Claude Platform, the model is available through Amazon Web Services, Google Cloud and Microsoft Azure.
The most notable new feature is the effort setting. Previously found only in larger models, this setting lets you get either fast, low-cost answers or answers that think more and cost more from the same model. The charts in the announcement show five levels: Low, Med, High, Xhigh and Max. According to the model documentation, the default level is medium.
| Feature | Haiku 5.5 |
|---|---|
| API ID | claude-haiku-5-5 |
| Context window | 1M tokens |
| Max output | 128K tokens |
| Knowledge cutoff | June 2026 |
| Default effort | Medium |
| Retirement | Not before October 7, 2027 |
How much does it cost?
The model's price has two tiers depending on request length. For requests up to 100K tokens, the price drops to one tenth of Haiku 4.5's. For longer requests, the discount is 50 percent.
| Per million tokens | Haiku 5.5 (≤100K / >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 read | $0.01 / $0.05 | $0.10 | $0.10 |
| Cache write | $0.125 / $0.625 | $1.25 | $2.50 |
One caveat: Haiku 5.5's updated tokenizer counts slightly more tokens for the same text, so the real savings vary from job to job. Anthropic's own estimate is a drop of around 75 percent per average job.
In the same announcement, the cache read price for Sonnet 5.5 was also cut from $0.20 to $0.10 per million tokens. Anthropic says this will lower Sonnet 5.5 costs by about 20 percent for most agent workloads.
Performance: how much better than Haiku 4.5?
The numbers show more than a one-generation gap between the two versions. Especially in computer use and terminal tasks, jobs that Haiku 4.5 almost never completed are now finished at a serious rate.
| Sonnet 5.5 | 83,9 |
|---|---|
| Haiku 5.5 | 72,4 |
| GPT-6 Luna | 48,9 |
| Haiku 4.5 | 15,7 |
Kaynak: Anthropic
| Benchmark | Haiku 5.5 | Haiku 4.5 | GPT-6 Luna | Sonnet 5.5 |
|---|---|---|---|---|
| GDPval-AA v2.1 (knowledge work) | 1620 | 735 | 1437 | 1840 |
| Terminal-Bench 4.0 | 39.2% | 0.0% | 16.4% | 70.6% |
| Humanity's Last Exam (with tools) | 57.4% | 18.7% | — | 64.5% |
| FrontierCode 1.1 (Main) | 46.4% | — | 42.4% | 52.1% |
Customer results point the same way. Asana reports more than 30 percent lower latency on task completion, while Box reports an 11-point higher score than Haiku 4.5 at roughly half the latency.
What is Claude Haiku 5.5 the right choice for?
Anthropic positions the model for narrowly scoped, high-volume work. According to the use cases listed in the announcement, the model stands out in the following jobs:
- Classification and routing. Distributing incoming support tickets, emails or forms to the right team.
- Summarization and compression. Turning long documents or agent conversations into short summaries.
- Data extraction. Reading structured fields from invoices, contracts or web pages.
- Live customer support. Chat interfaces where latency matters.
- Coding as a sub-agent. Small steps such as searching files or running tests inside a workflow orchestrated by Opus 5.5 or Sonnet 5.5.
For complex, multi-step software development, Sonnet 5.5 or Opus 5.5 remain the better choice. The gap between 39 percent and 71 percent on Terminal-Bench makes this plain.
What changed on the safety side?
Anthropic reports that Haiku 5.5 shows far less misaligned behavior than Haiku 4.5 and is less likely to go along with misuse. Its cybersecurity safeguards are stricter than Haiku 4.5's and looser than Sonnet 5.5's: defensive work is allowed more freely, while offense-oriented techniques such as penetration testing are blocked. Organizations that need broader access can apply to Anthropic's verification programs.
Frequently asked questions
How do I get started with Claude Haiku 5.5?
In the Claude API, just enter claude-haiku-5-5 as the model name. If you are migrating from Haiku 4.5, we recommend re-measuring your cost estimates with real traffic, since the token count changes.
I'm a Max or Team subscriber. Can I try Haiku 5.5 for free?
Indirectly, yes. The monthly Claude API credit announced the same day applies to all models, and thanks to Haiku 5.5's low price, that balance lasts longest on Haiku.
Does Haiku 5.5 replace Sonnet 5.5?
No. Haiku 5.5 is built for speed and cost. For complex reasoning and long coding tasks, Sonnet 5.5 and Opus 5.5 still give noticeably better results.
Can it be used through Bedrock or Vertex AI?
Yes. The model is also offered through Amazon Web Services, Google Cloud and Microsoft Azure.
Conclusion
Claude Haiku 5.5 makes a big leap in capability in the small-model class while cutting the price to one tenth. For any team with high-volume workloads, it is worth redoing the cost math. At EngerekTech, we choose models for client projects based on the type of work, and we share what we learn from AI integrations on this blog.
Sources
