Claude Haiku 5.5: Features, Pricing & Performance
Anthropic has completed its Claude 5.5 lineup with Claude Haiku 5.5, its smallest and fastest model in the new generation.
The biggest story is the price. Haiku 5.5 starts at just $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100K tokens.
At the same time, Anthropic has significantly improved its performance on agentic coding, computer use, document processing, and reasoning.
So, is Haiku 5.5 now good enough to replace larger models for everyday AI workloads?
What Is Claude Haiku 5.5?
Claude Haiku 5.5 is designed for high-volume and latency-sensitive AI applications.
It is particularly suitable for:
|
Unlike Haiku 4.5, Haiku 5.5 also brings features from Anthropic's larger 5.5 models, including adaptive thinking, configurable effort levels, a 1M-token context window, and up to 128K output tokens.
This makes it more capable while keeping it positioned as Anthropic's low-cost model.
Key Features
Much Lower Cost
For prompts up to 100K tokens, Haiku 5.5 costs significantly less than Haiku 4.5.
Adaptive Thinking
Developers can control how much reasoning the model uses through effort levels ranging from low to max.
This allows simple tasks to stay fast and inexpensive while more complex tasks can use additional reasoning.
Computer and Browser Use
Haiku 5.5 can work with computers and browsers, making it more practical for lightweight UI and browser automation.
A Strong Subagent
Anthropic also positions Haiku 5.5 as a worker model for Sonnet 5.5 and Opus 5.5.
A larger model can handle planning and decision-making while Haiku handles repetitive tasks such as reading documents, extracting information, and summarizing results.
Claude Haiku 5.5 Benchmark Results
Anthropic's results show a significant improvement over Haiku 4.5.
Benchmark | Haiku 5.5 | Haiku 4.5 | Sonnet 5.5 |
GDPval-AA v2.1 | 1620 | 735 | 1840 |
AA-Briefcase v1.1 | 1578 | 614 | 1824 |
OSWorld 2.1 | 72.4% | 15.7% | 83.9% |
Terminal-Bench 4.0 | 39.2% | 0% | 70.6% |
FrontierCode 1.1 | 46.4% | — | 52.1% |
Humanity's Last Exam | 45.9% | 10.2% | 56.9% |
Chartography | 46.4% | 6.4% | 61.6% |
The biggest improvement is in computer use and agentic work.
On OSWorld 2.1, Haiku 5.5 scores 72.4%, compared with just 15.7% for Haiku 4.5.
It also moves from 0% on Terminal-Bench 4.0 to 39.2%, showing that Haiku is becoming much more useful for coding agents.
Claude Haiku 5.5 Pricing
Per 1M Tokens | Up to 100K | Over 100K |
Input | $0.10 | $0.50 |
Output | $0.50 | $2.50 |
Cache read | $0.01 | $0.05 |
Cache write | $0.125 | $0.625 |
The Batch API provides an additional 50% discount on input and output.
One important detail: prompts above 100K tokens move into the higher pricing tier, so large-context applications should calculate costs carefully.
Which Claude Model Should You Use?
Model | Best For |
Haiku 5.5 | High-volume tasks, extraction, routing, subagents |
Sonnet 5.5 | Coding, everyday agents, complex workflows |
Opus 5.5 | Hard reasoning and demanding agentic tasks |
Fable 5.1 | Highest-end specialized workloads |
A simple rule is:
Haiku for volume. Sonnet for general work. Opus for difficult work.
When Should You Choose Haiku 5.5?
Choose Haiku 5.5 for:
|
Choose Sonnet or Opus instead when:
|
The goal is not to use the largest model everywhere. It is to use the cheapest model that can reliably complete the task.
Availability
Claude Haiku 5.5 is available through Anthropic's API and major cloud platforms, including:
Claude API
Amazon Bedrock
Google Cloud
Microsoft Foundry
The API model ID is:
claude-haiku-5-5
Conclusion
Claude Haiku 5.5 is a major improvement over Haiku 4.5, particularly for agentic tasks, computer use, and high-volume AI workloads.
Its biggest advantage is the combination of stronger performance and extremely low pricing.
For businesses processing large numbers of routine AI tasks, Haiku 5.5 could make advanced AI significantly more affordable.
The key question is no longer whether a small model can do the job. It's whether you really need a larger model for the job in the first place.
Comments