Claude Haiku 5.5: Anthropic’s Fastest, Cheapest AI Model

Anthropic has released Claude Haiku 5.5, describing it as the company’s cheapest, fastest, and most capable small Claude model to date.

The new model is designed for tasks where speed and low operating costs matter. That includes summarizing large volumes of documents, querying databases, handling customer-support conversations, and operating web browsers through AI agents.

For developers using Claude through Anthropic’s API, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts containing up to 100,000 tokens.

Tokens are the small pieces of text AI models process when reading and generating content. They are also the primary unit used by AI companies to calculate API costs.

The pricing puts Haiku 5.5 in direct competition with OpenAI’s GPT-6 Luna, which launched at the same $0.10-per-million-input-token rate.

Haiku 5.5 Cuts Costs by Up to 90%

The price difference from the previous generation is significant.

Claude Haiku 4.5 cost $1 per million input tokens and $5 per million output tokens. Haiku 5.5 drops those prices by 90%, making the new model considerably cheaper for high-volume applications.

Anthropic is also reducing the price for prompts containing more than 100,000 tokens by 50%.

The company says roughly 90% of requests made with the previous Haiku model were below the 100,000-token threshold. Because Haiku 5.5 also uses slightly more tokens to process the same amount of text, Anthropic estimates that users could see average savings of around 75%.

That makes the model particularly attractive for applications that process thousands or millions of relatively straightforward requests.

What Is Claude Haiku 5.5 Designed For?

Haiku has traditionally been positioned as Claude’s fast, lightweight model, and Haiku 5.5 continues that approach.

Customer-support conversations, document summaries, email processing, database queries, and other repetitive workflows are among its most obvious use cases.

In other words, Haiku 5.5 isn’t primarily aimed at replacing Anthropic’s most advanced models for difficult reasoning. Instead, it is built for situations where an AI system needs to respond quickly and economically at scale.

That distinction matters for companies running AI-powered products. If an application has to handle thousands of customer interactions every hour, even a small reduction in the cost of each request can have a major impact on the overall bill.

Haiku 5.5 Is More Capable Than Its Price Suggests

Low cost does not mean Haiku 5.5 is a basic model.

Several benchmarks show that it can handle demanding computer-use and coding tasks, particularly when compared with other smaller models.

On OSWorld 2.1, which measures how well AI systems can operate a real computer through multi-step tasks, Haiku 5.5 achieved a 72.4% success rate. OpenAI’s Luna scored 48.9%.

The gap was even larger on Terminal-Bench 4.0. The benchmark gives AI agents professional tasks that require them to independently execute commands in a terminal and measures how often they complete those tasks correctly on the first attempt.

Haiku 5.5 scored 39.2%, compared with 16.4% for OpenAI’s Luna and 0% for Haiku 4.5.

Anthropic’s larger Claude Sonnet 5.5 scored 70.6% on the same benchmark, showing that Haiku 5.5 is still positioned below the company’s more capable models while delivering a much lower operating cost.

Fast Responses Don’t Guarantee Correct Answers

Speed is one of Haiku 5.5’s biggest strengths, but users should not confuse fast output with reliable reasoning.

In a simple logic test, the model responded almost instantly. However, it produced an incorrect answer.

That example highlights an important trade-off with smaller, speed-focused models. Haiku 5.5 can be extremely useful for high-volume workflows, but important decisions should still be checked rather than accepted automatically.

For applications where accuracy is more important than response time or cost, Anthropic’s larger models may remain the better option.

Strong Results on Professional Work

Haiku 5.5 also performed well on GDPval-AA v2.1, a benchmark designed to evaluate AI models on professional tasks across 44 different occupations.

The benchmark uses an Elo-style rating system, similar to the ranking method commonly associated with chess.

Haiku 5.5 achieved a score of 1,620. OpenAI’s Luna scored 1,437, while Haiku 4.5 scored 735.

The results suggest that Anthropic has improved Haiku considerably compared with its previous generation, particularly when the model is used for practical professional workflows.

Adjustable Effort Comes to Haiku

Haiku 5.5 introduces another feature that was previously associated with Anthropic’s more advanced models: an adjustable effort setting.

The setting allows users to decide how much computational effort the model should use when generating an answer.

That creates a practical trade-off. Users can prioritize speed and lower costs for simple tasks, while allowing the model to spend more effort when a request requires deeper reasoning.

Anthropic is also reducing the cache-read price for Sonnet 5.5 to $0.10 per million tokens. Cached tokens refer to text the model has already processed, allowing applications to reuse previously handled information at a lower cost.

Haiku 5.5 Completes Anthropic’s Claude 5.5 Lineup

Haiku 5.5 arrives 15 days after Claude Opus 5.5, which launched on September 22, and nine days after Sonnet 5.5, released on September 28.

With Haiku 5.5, Anthropic has now completed the three-model Claude 5.5 lineup it previously promised.

Opus 5.5 was the first of the three releases and came shortly after Anthropic CEO Dario Amodei published an essay calling on the AI industry to slow the pace of capability improvements.

The three models serve different purposes within Anthropic’s lineup: Opus focuses on the company’s most demanding workloads, Sonnet sits between performance and cost, while Haiku emphasizes speed and affordability.

Where Can You Use Claude Haiku 5.5?

Claude Haiku 5.5 is now available through the Claude website as well as Amazon Web Services, Google Cloud, and Microsoft Azure.

For API integrations, the model is available under the name claude-haiku-5-5.

Anthropic is also introducing monthly API credits for some subscription plans. Subscribers on the Max 5x plan receive $100 in credits, while Max 20x subscribers receive $200. Team plans can receive up to $500 in pooled credits across users.

Final Thoughts

Claude Haiku 5.5 is not simply a cheaper version of Anthropic’s larger models. Its biggest advantage is the combination of speed, low cost, and improved capability.

The model appears particularly well suited to companies that need to process large numbers of requests without paying premium prices for every interaction. Customer support, document processing, browser automation, database queries, and other repetitive AI workloads could benefit significantly from the lower pricing.

Its benchmark performance also shows that Anthropic has pushed Haiku beyond the role of a basic lightweight model. However, its occasional reasoning mistakes are a reminder that speed and affordability still need to be balanced against accuracy.

For developers building AI products at scale, that balance could make Haiku 5.5 one of the more interesting small-model releases in Anthropic’s Claude lineup.