Claude Haiku 4.5

What it is

Claude Haiku 4.5 is Anthropic’s current small, fast and low-cost model, distributed under the identifier claude-haiku-4-5. Across the Claude family, Haiku is the label Anthropic gives to its most compact tier: the model built to answer quickly and cheaply rather than to top every reasoning benchmark. Claude Haiku 4.5 is the latest to carry that name, and it takes over the role its predecessors held, handling the high-volume, latency-sensitive work that would be wasteful to route through a lab’s largest and priciest systems.

Why a fast, cheap tier

The case for a fast, cheap tier is straightforward. Most production systems do not need a flagship model on every request. Sorting support tickets, pulling fields out of documents, tagging content, drafting short replies, checking a passage against a rule, or running the first pass of a longer pipeline: these are jobs where speed and price per token count for more than the final few points on a hard exam. Claude Haiku 4.5 is designed to sit in exactly that slot, returning answers rapidly and at a fraction of the cost of a top-tier model, so the expensive tier can be kept for the smaller share of requests that truly need it.

That division of labour has become the standard way to build with large language models, and it is worth understanding why a model like Claude Haiku 4.5 is so often the workhorse rather than the headline act. When a task runs thousands or millions of times a day, the difference between a cheap model and a flagship is the difference between a viable product and an unaffordable one. A compact model that responds in a fraction of a second also makes an interface feel immediate, which is why the fast tier tends to power autocomplete, live classification, routing and the many small calls hidden inside agentic workflows.

Where 4.5 sits

The “4.5” in the name signals where it sits in Anthropic’s release cadence. Each turn of the Haiku line has tended to narrow the gap with the larger tiers while holding on to the speed and low price that define the tier, so buyers can often move to a newer Haiku and get better answers at a similar cost. Anthropic exposes the model to developers through its API, and the practical details that decide whether it fits a given job, its context window, its pricing per million tokens, its supported features and its precise capabilities, are published in Anthropic’s own documentation at docs.claude.com rather than restated second hand here.

The wider pattern

For readers new to the way these families are structured, Claude Haiku 4.5 is a useful illustration of a wider pattern. Frontier labs no longer ship a single model; they ship a range, from a small fast tier through to a large flagship, and expect developers to route each request to the cheapest model that can do the job well. Haiku is Anthropic’s answer at the small end. It is rarely the model you reach for when a task demands the deepest reasoning, yet it is frequently the right default for everything else, and getting that routing decision right is one of the main levers for controlling cost in any serious deployment.

What to watch

Where this goes next is a question of cadence rather than surprise. Anthropic has kept Haiku, and the family it belongs to, on a steady release schedule, so the sensible thing to watch is the next revision and how far it closes the distance to the larger tiers while staying fast and cheap. For anyone building on Anthropic today, Claude Haiku 4.5 is the current name to know at the economical end of the range. For the broader picture of how these systems are organised and compared, see our large language models hub and the wider AI section.