Claude Haiku 5.5 cuts API prices by up to 90%
Anthropic's new small model costs $0.10 per million input tokens on short requests, and Max and Team plans now get monthly API credits.
Updated
In short
- 01Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens on requests up to 100,000 tokens.
- 02Anthropic says the average Haiku job now costs about 75% less, since 90% of Haiku 4.5 requests were under that limit.
- 03Claude Max and Team subscribers get $100 to $500 a month in API credits, rolling out this week.
The story
Anthropic released Claude Haiku 5.5 on October 7, 2026. Its launch post calls it the cheapest and fastest small model the company has made. It's designed for high-volume tasks like summaries, classification and database queries, and for working as a subagent under the bigger Claude models.
It's the third Claude 5.5 model in a month, Reuters reported, as Anthropic expands its lineup ahead of a planned IPO.
Haiku 5.5 is up to 90% cheaper than Haiku 4.5
Requests up to 100,000 tokens cost $0.10 per million input tokens and $0.50 per million output tokens, down from $1 and $5 on Haiku 4.5. Longer requests cost $0.50 and $2.50, half the previous price.
Anthropic says about 90% of Haiku 4.5 requests fell under that threshold. It estimates the average saving at about 75%, because a new tokenizer uses slightly more tokens to complete the same task.
On short requests, Haiku now matches the price of OpenAI's GPT-6 Luna, according to VentureBeat's price comparison. Above 100,000 tokens, Haiku costs five times as much as Luna, whose higher rate only begins at 272,000 input tokens.
Anthropic's benchmark scores are strong, but early tests are mixed
Haiku 5.5 is the first Haiku model with an adjustable effort setting, from low to max. In Anthropic's own evaluations, it scored 72.4% on the OSWorld 2.1 computer-use benchmark, compared with 15.7% for Haiku 4.5.
Those are vendor-reported results, not independent ones. VentureBeat pointed out that the 39.2% Terminal-Bench coding score comes from maximum effort, while the default medium setting scores about 20%. Anthropic still recommends Sonnet 5.5 and Opus 5.5 for complex coding.
The Hacker News launch thread drew 488 comments. Of the top 30 on October 11, about a third praised the price or speed, and about one in six reported weak or mixed results. That's an early snapshot of developer reactions, not a survey.
I was curious if this smaller model was good enough for complex UI. It was not.
jjcm on Hacker News, October 7
Max and Team subscribers get monthly API credits
Anthropic also halved the price of cache reads on Sonnet 5.5, to $0.10 per million tokens. Starting this week, Max 5x subscribers receive $100 a month in API credits, Max 20x subscribers $200 and Team plans up to $500.
Haiku 5.5 is available now on the Claude Platform as claude-haiku-5-5, and through Amazon Web Services, Google Cloud and Microsoft Azure.
Why it matters
Small models handle the routine work inside AI agents, and price usually decides which one developers choose. Matching Luna on short requests puts Haiku back in contention. The catch is the 100,000-token threshold: agent workloads cross it quickly, and beyond it Haiku costs five times as much as Luna.
Tools in this story
Share
