Claude Haiku 5.5 Launches With Lower API Prices and Adjustable Effort

Anthropic has introduced Claude Haiku 5.5, a smaller AI model designed to make routine, high-volume tasks faster and less expensive. Announced on October 7, 2026, it is available through Claude and the company's developer platform, as well as supported cloud providers. The release matters less as a headline-grabbing race for the largest model and more as a practical change for teams that need thousands or millions of short AI requests to be economical.

What is new in Haiku 5.5?

Haiku is the speed-and-efficiency tier in Anthropic's Claude family. According to the company's announcement, Haiku 5.5 is its fastest standard-speed model and its most capable Haiku to date. Anthropic positions it for classification, summarization, document lookups, customer support, simple coding changes and subagent tasks that a larger model can delegate. That does not mean Haiku replaces Claude's more capable models for complicated reasoning or extended engineering projects. The company's own comparison still places Sonnet 5.5 and Opus 5.5 ahead on demanding agentic coding evaluations.

One meaningful addition is adjustable effort. Haiku 5.5 is the first Haiku-class release to let developers tune how much work the model performs for a request. In principle, that gives an application a way to spend less on straightforward jobs and allow more effort when a task needs it. The right setting will depend on testing with the application's actual prompts rather than assuming one configuration fits everything.

Pricing: why the smaller model matters

Anthropic lists Haiku 5.5 API pricing for prompts up to 100,000 tokens at approximately ₱6.27 (US$0.10) per million input tokens and ₱31.37 (US$0.50) per million output tokens. For prompts above 100,000 tokens, the listed rates rise to approximately ₱31.37 (US$0.50) for input and ₱156.86 (US$2.50) for output per million tokens. The peso amounts are indicative conversions using approximately ₱62.745 per US$1 on October 10, not official Philippine prices; cloud-provider charges, taxes, and exchange rates may differ.

For comparison, the previous Haiku 4.5 model was listed at ₱62.75 (US$1.00) per million input tokens and ₱313.73 (US$5.00) per million output tokens. Anthropic says its new model is priced 90% lower for requests within the shorter-prompt tier, though its real-world cost reduction estimate also considers differences in token usage. The important distinction is that the low headline price is conditional on prompt size. Teams using long documents or large histories should calculate their own bills before switching.

The company also reduced cache-read pricing for Sonnet 5.5 from roughly ₱12.55 (US$0.20) to ₱6.27 (US$0.10) per million tokens. Anthropic estimates that change cuts the cost of many agent workflows by around 20%. It is a separate pricing change, not a claim that Sonnet and Haiku now cost the same.

What developers can actually build

A Philippine startup handling customer inquiries could use a smaller model to classify incoming tickets, extract structured fields, summarize a case and hand uncertain results to a person or a larger model. A developer could use Haiku as a quick subagent for a tightly scoped edit while keeping more complex architectural decisions with a stronger model. These are possible applications, not independent performance tests or promises of accuracy.

Anthropic also highlights computer and browser use, along with updates to its Python and TypeScript SDKs for those capabilities in beta. Businesses should still test permission boundaries, review error rates and keep human oversight where incorrect actions could affect customers or records. Faster responses are only useful when the output is sufficiently reliable for the workflow.

The launch includes availability through Anthropic's platform and major cloud services. On Anthropic's own API, the model identifier is `claude-haiku-5-5`. Claude users can also select Haiku 5.5 in supported experiences, although access and limits may vary by account or platform. No Philippines-specific retail subscription price was announced in the model release.

What to watch next

The clearest test is not whether Haiku 5.5 can top a single benchmark. It is whether developers can move common jobs to a smaller model without raising the cost of quality checks, retries or human correction. Anthropic published benchmark comparisons and early customer evaluations, but these are vendor-supplied figures; teams should verify performance on their own tasks.

For readers in the Philippines, the practical takeaway is straightforward: a cheaper API tier can make useful AI features easier to deploy in local apps, support tools and coding workflows. The actual savings depend on token volumes, prompt lengths, hosting choices and the value of a correct answer. The model is available now, but any local subscription or billing terms should be checked directly before purchase.

Source: Anthropic's official Haiku 5.5 announcement and model overview.