Anthropic Launches Claude Haiku 5.5 at 75 Percent Lower Running Cost

Why a cheaper small model matters more than another flagship: Haiku 5.5 changes the maths for every routine AI call a business makes

October 8, 2026 — Anthropic released Claude Haiku 5.5 on October 7, positioning it as its cheapest, fastest and most capable small model yet. The company says it costs around 75 percent less to run on average than Haiku 4.5, while adding a larger context window and a new control for balancing cost against intelligence.

Image: A lightweight AI model routing high-volume tasks through a data centre. Credit: AI-generated illustration / Red Eye Eagle News, original illustration.

Editor's Highlight: The story is not raw power. Haiku 5.5 is built for repetitive work that dominates real API bills — sorting, summarising and checking — and its short-prompt price now matches OpenAI's GPT-6 Luna, turning the small-model tier into a direct price contest.

Table of Contents

• Key Facts
• A Cheaper Small Model, Not Just a Price Cut
• Pricing in Two Tiers
• What the Benchmarks Show
• Built for High-Volume and Subagent Work
• Wider Claude Changes Alongside the Launch
• What Developers Should Check Before Switching
• Bottom Line
• FAQ
• Sources

Key Facts

Detail	Information
Company	Anthropic
Model	Claude Haiku 5.5, ID claude-haiku-5-5
Release date	October 7, 2026
Short-prompt price	$0.10 input / $0.50 output per million tokens, up to 100,000 tokens
Long-prompt price	$0.50 input / $2.50 output per million tokens, above 100,000 tokens
Context window	1 million tokens, up from 200,000 in Haiku 4.5
Max output	128,000 tokens
Availability	Claude Platform, AWS, Google Cloud, Microsoft Azure

A Cheaper Small Model, Not Just a Price Cut

Haiku is the smallest tier in the Claude family, below Sonnet and Opus. Anthropic says it is the third model in the Claude 5.5 family, following Opus 5.5 and Sonnet 5.5 earlier this fall.

Haiku 4.5, released about a year ago, charged $1 per million input tokens and $5 per million output tokens. Against that baseline, the new short-prompt rate is about a 90 percent reduction.

The average saving is lower, Anthropic says, once longer prompts, caching and a revised tokenizer are taken into account.

Newsroom Focus: Small models rarely lead headlines, yet they handle most production traffic. A price move here affects far more calls than a flagship launch aimed at complex reasoning alone.

Pricing in Two Tiers

The price depends on prompt length. Keeping a prompt at or below 100,000 tokens keeps it in the cheaper band — a band Anthropic says covered around 90 percent of requests to the previous Haiku model.

Model / band	Input per 1M	Output per 1M
Haiku 5.5, up to 100K	$0.10	$0.50
Haiku 5.5, above 100K	$0.50	$2.50
Haiku 4.5	$1.00	$5.00
GPT-6 Luna, short prompts	$0.10	$0.50

In the short band, cache reads are listed at $0.01 per million tokens and cache writes at $0.125, according to launch pricing reports.

Token counting also matters. Several reports note the new tokenizer counts the same text as more tokens than the old model did, so teams should measure their own prompts rather than assume the list-price cut flows through unchanged.

What the Benchmarks Show

All figures below are Anthropic-reported results, not independent tests. They should be read as the company's claims until other evaluators publish their own runs.

On the offline subset of OSWorld 2.1, a computer-use test, Haiku 5.5 scored 72.4 percent, compared with 15.7 percent for Haiku 4.5 and 48.9 percent for GPT-6 Luna in the same reported table.

On Terminal-Bench 4.0 it scored 39.2 percent, where Haiku 4.5 scored zero. On FrontierCode 1.1 it reached 46.4 percent, compared with 42.4 percent for Luna, and on Humanity's Last Exam it scored 45.9 percent without tools and 57.4 percent with tools.

Analysis Note: The largest jump is in computer use, not general chat. That fits the intended role: a fast model that can operate software and complete short, well-defined steps inside a larger workflow. Anthropic still directs complex agentic coding toward Sonnet 5.5 and Opus 5.5.

Built for High-Volume and Subagent Work

Anthropic describes the model as designed for high-volume, cost-sensitive tasks. Examples across the launch coverage are consistent: summaries, context compaction, database queries and classification requests.

Speed-sensitive work is a second fit, including live customer support and browser use, where a slower response is visible to the end user.

A third role is as a subagent. A larger model plans and handles difficult steps, while Haiku handles the frequent, narrower calls in between.

The model takes text and image input and produces text output. It is also the first Haiku-class model with an adjustable effort setting, letting a developer trade lower cost against higher intelligence instead of switching tiers.

Practical Takeaway: Start with one repetitive workflow, such as ticket routing or document extraction. Compare cost, latency and accuracy there before moving customer-facing traffic.

Wider Claude Changes Alongside the Launch

Anthropic said it is halving the price of Sonnet 5.5 cache reads, a change the company says makes Sonnet about 20 percent cheaper on most agentic work.

The company also introduced a new monthly API credit for Claude Max and Team subscribers, intended to help them build agents and applications on the Claude Platform.

What Developers Should Check Before Switching

First, check prompt length. A workflow that grows past 100,000 tokens per request moves into the higher band, at five times the short-band rate.

Second, check token counts under the new tokenizer on real traffic, and test the effort setting at more than one level. Lower effort may suit simple classification, while longer tasks may need a higher setting.

Finally, keep the larger models in the comparison. Haiku is positioned for narrow jobs run repeatedly at scale, not as a replacement for Sonnet or Opus on complex reasoning.

Bottom Line

Claude Haiku 5.5 makes routine AI work much cheaper on paper, expands context to 1 million tokens and adds an effort control the Haiku tier did not have before. Watch independent testing and any rival price response next.

FAQ

When was Claude Haiku 5.5 released?

Anthropic released it on October 7, 2026. It is the third model in the Claude 5.5 family, after Opus 5.5 and Sonnet 5.5.

How much does Haiku 5.5 cost?

For prompts up to 100,000 tokens, $0.10 per million input tokens and $0.50 per million output tokens. Above that threshold, $0.50 and $2.50.

What is Haiku 5.5 best used for?

According to Anthropic, high-volume and speed-sensitive work such as summaries, classification, database queries, live support, browser use and subagent tasks.

How large is its context window?

1 million tokens, up from 200,000 in Haiku 4.5. Maximum output is 128,000 tokens.

Where can developers use it?

On the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure, under the model ID claude-haiku-5-5.

Sources

• TechStartups — Oct 8 roundup, Haiku 5.5: https://techstartups.com/2026/10/08/top-tech-news-today-october-8-2026-globalfoundries-google-manus-microsoft-nvidia-openai-tencent-more/
• Startup Fortune — Haiku 5.5 launch, tiers and availability: https://startupfortune.com/anthropic-launches-claude-haiku-55-with-prices-cut-up-to-90-percent/
• 9to5Mac — Haiku 5.5 release and credits: https://9to5mac.com/2026/10/07/anthropic-upgrades-claude-with-new-haiku-5-5-model-details-here/
• TestingCatalog — Haiku 5.5 prices and benchmarks: https://www.testingcatalog.com/anthropic-launches-claude-haiku-5-5-at-lower-api-prices/
• MarkTechPost — Haiku 5.5 pricing and benchmarks: https://www.marktechpost.com/2026/10/07/anthropic-releases-claude-haiku-5-5-a-small-model-with-1m-context-priced-at-0-10-per-million-input-tokens/
• AI Agent News — Oct 8 roundup, Haiku 5.5: https://aiagentstore.ai/ai-agent-news/today
• Image source — AI-generated illustration, Red Eye Eagle News.