How much cheaper is Haiku 5.5 than Sonnet 5.5?

In the test, Haiku 5.5 averaged about 1/17th the cost of Sonnet 5.5 on the medium setting. On Anthropic's published API pricing, Sonnet 5.5 costs 20 times more than Haiku 5.5.

Is GPT-6 Luna cheaper than Haiku 5.5?

Yes. GPT-6 Luna was cheaper and faster in these tests, and its API price is the same as Haiku 5.5.

Why did Haiku 5.5 score low on the English email task?

On the medium setting, it failed twice out of three times by misreading its own negotiating position. Raising the thinking setting to max resolved this but took 254.1 seconds and 4.68 yen per response.

Large Language ModelsITmedia AI＋3h agoOct 16, 2026, 13:00 JST

Claude Haiku 5.5: 1/17 the cost, but GPT-6 Luna is cheaper

Send on LINE

3 Key Points

What happened
In the test, Haiku 5.5 on medium setting averaged scores close to Sonnet 5.5 on routine documents at about 1/17th the cost, but dropped to 6.3 points on the English email task.
Why it matters
The test backs Anthropic's claim that Haiku 5.5 suits cost- and speed-focused, narrow tasks, but for complex document work a Sonnet-tier model looks more suitable.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

Anthropic released Claude Haiku 5.5 on October 7, positioning it as its lightest model suited to cost- and speed-focused tasks. The company itself limits its recommended uses to summaries, context compression, and sub-agents, saying Sonnet 5.5 or Claude Opus 5.5 remain better for complex agentic coding.

The test pitted Haiku 5.5 against Claude Sonnet 5.5 and GPT-6 Luna across three tasks: an apology email for a late delivery, meeting minutes from handwritten notes, and summarizing and replying to an English email from an overseas client. Scoring was done by Claude Fable 5.1 and GPT-6 Astra, averaged and rounded. Results showed Haiku 5.5 performed well on routine documents but struggled when it needed to grasp its own position.

Artificial Analysis's Intelligence Index placed Haiku 5.5 (max) at 43 points, up 26 points from Claude Haiku 4.5 and above GPT-6 Luna (38), but noted it outputs about three times as many tokens per task as GPT-6 Luna's max. Anthropic's API pricing shows Haiku 5.5 at $0.10 input and $0.50 output per 1 million tokens for prompts under 100,000 tokens — matching GPT-6 Luna and 20 times cheaper than Sonnet 5.5.

FAQ

How much cheaper is Haiku 5.5 than Sonnet 5.5?: In the test, Haiku 5.5 averaged about 1/17th the cost of Sonnet 5.5 on the medium setting. On Anthropic's published API pricing, Sonnet 5.5 costs 20 times more than Haiku 5.5.
Is GPT-6 Luna cheaper than Haiku 5.5?: Yes. GPT-6 Luna was cheaper and faster in these tests, and its API price is the same as Haiku 5.5.
Why did Haiku 5.5 score low on the English email task?: On the medium setting, it failed twice out of three times by misreading its own negotiating position. Raising the thinking setting to max resolved this but took 254.1 seconds and 4.68 yen per response.

ITmedia AI＋Read Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.