TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Anthropic has launched Claude Haiku 5.5, a small model it describes as its fastest to date and designed for high-volume, cost-sensitive tasks. The company says it costs about 75% less to run than Haiku 4.5 on average, while published benchmark results show it trailing larger Claude models on some complex tasks. Haiku 5.5 is available now across Anthropic’s platform and major cloud providers.
Anthropic has released Claude Haiku 5.5, a small model aimed at high-volume and speed-sensitive tasks, and says it costs about 75% less to run on average than Haiku 4.5. The launch matters to developers building AI applications because it offers a lower-cost option for recurring workloads, though Anthropic’s benchmarks show Haiku 5.5 remains behind its larger models on some demanding tasks.
Anthropic says Haiku 5.5 is built for tasks such as summarization, prompt compaction, database queries and classification, as well as live customer support and browser use. The company calls it its fastest model to date and says it can serve as a subagent alongside Sonnet 5.5 or Opus 5.5 in coding workflows. Those descriptions are Anthropic’s product claims; the company’s published evaluations provide benchmark comparisons, not independent confirmation of performance in every customer setting.
For prompts up to 100,000 tokens, Anthropic lists Haiku 5.5 prices of $0.10 per million input tokens and $0.50 per million output tokens. For prompts above that length, the listed rates rise to $0.50 for input and $2.50 for output per million tokens. Anthropic says prompts up to 100,000 tokens represented around 90% of requests to its previous Haiku model. Its stated average cost reduction of about 75% is a comparison with Haiku 4.5, not a guarantee that every task will cost that much less.
The release includes an adjustable effort setting, which lets users choose between lower cost and more reasoning effort, according to Anthropic. The company also cut Sonnet 5.5 cache-read prices by 50%, to $0.10 per million tokens, and says the change makes Sonnet 5.5 about 20% cheaper on most agentic work. Haiku 5.5 is available through the Claude Platform and on Amazon Web Services, Google Cloud and Microsoft Azure.
Lower Costs for Routine AI Work
The launch targets a practical constraint for businesses deploying AI agents: the cost and speed of handling frequent, bounded tasks. A lower-priced model could make it more economical to summarize long interactions, classify incoming requests or carry out preliminary steps before a more capable model handles a difficult problem. Anthropic positions Haiku 5.5 for this kind of division of labor, rather than as a replacement for every model in its range.
That distinction matters because model choice affects both expense and task performance. In Anthropic’s published results, Haiku 5.5 scored 39.2% on Terminal-Bench 4.0, compared with 70.6% for Sonnet 5.5, a benchmark the company describes as measuring complex, multistep command-line work. The figures are vendor-reported evaluations, and benchmark scores do not by themselves establish how a model will perform in a particular product. Still, they support Anthropic’s recommendation to reserve larger models for more complex agentic coding.
Anthropic also reported early customer feedback from Asana. Staff software engineer Aaron Vinh said the company saw more than 30% lower latency for task completion and up to 2.5 times faster inference per agent turn in its internal evaluation suite. That result is an Asana account of its own testing, not a general performance guarantee. It illustrates why speed may matter alongside token prices: quicker responses can improve the feel of interactive tools, while actual savings and quality depend on a customer’s workload.
As an affiliate, we earn on qualifying purchases.
Where Haiku Fits in Claude
Haiku is Anthropic’s smaller-model tier, positioned for fast, frequent requests, while Sonnet and Opus are intended for more demanding work. Anthropic’s launch materials describe Haiku 5.5 as suitable for narrow tasks and for use as a subagent, but say Sonnet 5.5 and Opus 5.5 remain better suited to complex agentic coding. The intended setup is to match model capability to each step rather than route every request to the largest model.
The published comparison includes several evaluations, among them computer use, knowledge work, multidisciplinary reasoning and agentic coding. Results vary by test: Haiku 5.5 records 72.4% on the OSWorld 2.1 offline subset, compared with 83.9% for Sonnet 5.5, while its Terminal-Bench score is well below Sonnet’s. Anthropic points readers to the model’s system card for evaluation methods and safety details. These results should be read as company-reported benchmark measurements, not a complete measure of real-world reliability.
The company says Haiku 5.5 showed improvements over Haiku 4.5 in almost all of its alignment evaluations, with fewer instances of misaligned behavior and less willingness to cooperate with misuse. Anthropic also describes cybersecurity safeguards that block penetration testing and other techniques it considers more likely to be used by attackers, while allowing a wider range of defensive tasks than its safeguards for Sonnet 5.5. These are Anthropic’s descriptions of its safety testing and controls; the system card contains the underlying details.
““the cheapest, fastest, and most capable small model we’ve ever released””
— Anthropic
cloud-based AI model deployment services
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Published Evidence
Anthropic’s announcement does not provide an independent evaluation of Haiku 5.5, and the supplied material does not state a release calendar date. Its stated cost reduction is an average; actual spending will depend on prompt length, token use, caching and task design. The pricing table also sets different rates for prompts above and below 100,000 tokens, so customers need to check which tier applies to their usage.
It is not yet clear how Haiku 5.5 will perform across a broad range of production deployments or how often users will need to hand difficult tasks to a larger model. Asana’s latency report reflects one company’s evaluation and comparison model. Anthropic’s benchmark scores and safety findings likewise come from the company’s own evaluation process, described in its system card; they do not settle how the model will behave in every real-world setting.
As an affiliate, we earn on qualifying purchases.
Availability and Developer Evaluation
Anthropic says Haiku 5.5 is available now through its Claude Platform, using the model identifier claude-haiku-5-5, as well as through Amazon Web Services, Google Cloud and Microsoft Azure. Developers can consult the company’s migration guide and system card for integration and evaluation details. The announcement does not give a separate rollout timetable for those platforms.
The next useful test for customers is whether the lower listed rates and Anthropic’s speed claims hold for their own prompts and workflows. Developers can compare Haiku 5.5 with their current model on task accuracy, latency and total cost, particularly when a system routes difficult requests to Sonnet 5.5 or Opus 5.5. Anthropic also said it is introducing a monthly API credit for Claude Max and Team subscribers to support development on its platform, but the source does not specify the credit amount or full terms.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Claude Haiku 5.5?
Claude Haiku 5.5 is Anthropic’s small model for fast, high-volume tasks such as summarization, classification and database queries. Anthropic says it can also be used as a subagent in coding workflows.
How much cheaper is it than Haiku 4.5?
Anthropic says Haiku 5.5 costs about 75% less to run on average than Haiku 4.5. The actual cost depends on usage, and listed token prices vary for prompts above 100,000 tokens.
Where can developers access Haiku 5.5?
Anthropic says it is available now on the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure. The Claude Platform model identifier is claude-haiku-5-5.
Is Haiku 5.5 intended for complex coding tasks?
Anthropic presents it as useful for narrower coding tasks and as a subagent, but says Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding. Its published Terminal-Bench 4.0 score is below Sonnet 5.5’s.
Source: hn
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
