Anthropic’s Latest Haiku Release Cuts Inference Cost 75%

Anthropic released Claude Haiku 5.5 on Wednesday, pricing its newest small model at 75% less to run than the prior Haiku 4.5 generation.

Key Takeaways

Anthropic released Claude Haiku 5.5 on Wednesday at 75% less cost to run than the prior Haiku 4.5 generation
Haiku is Anthropic’s smallest model tier for high-volume, latency-sensitive tasks including customer support bots and simple coding completions
The 75% reduction applies to input and output tokens processed when developers run the model at scale
As of Wednesday evening, Anthropic had not published pricing tables or a dedicated blog post for Haiku 5.5

The company says the reduction makes it the fastest and cheapest model it has shipped to date.

Haiku is its smallest model tier, built for high-volume, latency-sensitive tasks such as customer support bots and simple coding completions rather than complex reasoning.

The 75% cut applies to running the model at scale, developers pay less per token processed, whether input or output.

Industry Shifts Toward Cheaper Small Models

Anthropic, OpenAI, and Google have each released cheaper small models through 2026 as enterprise customers push back on inference bills tied to scaling chatbot and agent deployments.

A 75% cost cut on a tier used for repetitive, high-volume calls lowers the operating cost of agentic workflows, where a single task can trigger dozens of model calls in sequence.

Also Read: Claude Models Unlock Access For Approved Cyber Defenders

It demonstrates a lower cost claim, not ordinary-day performance, the provided release information does not state availability, pricing tables, rate limits, regions, or tiers.

A History Built On Budget Tiers

Haiku models have existed since the original Claude 3 family, positioned as the budget tier beneath Sonnet and Opus. Each iteration has prioritized latency and cost over raw capability, mirroring OpenAI’s mini and nano model lines.

Haiku 5.5 is described as the first Haiku model with a new, unspecified capability beyond the cost cut.

Details Still Missing From Launch

The move follows an industry pattern this year, flagship models become more expensive to train while the cheapest usable tier gets cheaper to run. As of Wednesday evening, the company had not published pricing tables or a dedicated blog post for Haiku 5.5.

An official announcement with per-token pricing and benchmark scores would confirm the reduction’s scale.