Anthropic has released Claude Haiku 5.5 — a new version of the fastest and cheapest model in the Claude lineup. On short prompts the price dropped 10x compared with Haiku 4.5, and benchmark scores jumped several times over. But the low price comes with two catches that most news stories skip. Here's what Haiku 5.5 really costs and where affiliates can use it.
How much Claude Haiku 5.5 costs
Prices per 1M tokens from Anthropic's official pricing:
- Input: $0.10 (Haiku 4.5: $1);
- Output: $0.50 (was $5);
- Cache reads: $0.01 (was $0.10);
- Batch API: input $0.05, output $0.25.
Catch 1: long prompts cost 5x more
The $0.10 / $0.50 price only applies while a prompt is up to 100,000 tokens. Above that, input is $0.50, output $2.50, and caching is also 5x pricier. That's why Anthropic itself says the model is "around 75%" cheaper than Haiku 4.5 on average: 90% savings on short prompts, 50% on long ones.
Catch 2: more tokens
Haiku 5.5 uses a newer tokenizer, and according to Anthropic's docs the same text takes about 30% more tokens than on Haiku 4.5. Factoring that in, the real savings on the same tasks are:
- short prompts (up to 100,000 tokens) — about 87%;
- long prompts — about 35%.
One more thing: adaptive thinking is on by default, and thinking tokens are billed as output. For simple tasks you can turn it off — at high effort or below.
What's new
- Effort setting. The first Haiku-class model with adjustable effort: low, medium, high, xhigh or max — you decide what matters more, price or answer quality.
- 1M-token context and up to 128,000 output tokens.
- Big benchmark gains. Computer use (OSWorld 2.1): 72.4% vs 15.7% for Haiku 4.5. Humanity's Last Exam with tools: 57.4% vs 18.7%. Terminal-Bench 4.0: 39.2% vs 0%.
- Browser use tool on the Claude API and Google Cloud.
What it's for
Anthropic positions Haiku 5.5 for high-volume, low-cost work: summarization, classification, database queries, live customer support, browser use and acting as a subagent alongside Opus 5.5 and Sonnet 5.5. For complex coding the company still recommends Sonnet 5.5 and Opus 5.5.
How to use it in affiliate marketing
- Support and warm-up bots. Replies in a Telegram bot or landing-page chat: the model is fast and cheap. 1,000 conversations with 2,000 input and 300 output tokens each cost about $0.46 vs $3.5 on Haiku 4.5 — not counting thinking.
- Sorting leads and comments. Classify applications, filter junk and fraud patterns, tag reviews and comments under creatives.
- Bulk copy. Headline and description variants and localization by GEO — cheap to generate dozens of versions for tests.
- Data digests. Summaries of reports, tracker logs and exports — in short prompts, to stay out of the expensive tier.
More AI tools for the job: AI tools for affiliate marketing.
Important if you already run bots on Haiku 4.5
When switching to Haiku 5.5, some old settings break requests:
- non-default
temperature,top_pandtop_kreturn an error — remove them; - assistant message prefill no longer works — messages must end with a user turn;
- manual thinking via
budget_tokensis replaced by adaptive thinking; - recheck
max_tokenslimits and cost estimates — the new tokenizer means more tokens.
The API model ID is claude-haiku-5-5. According to Anthropic, the model is available on the Claude API and on AWS, Google Cloud and Microsoft Azure.
In short
- Claude Haiku 5.5 came out on October 7: $0.10 input and $0.50 output per 1M tokens.
- Prompts over 100,000 tokens cost 5x more.
- The same text takes about 30% more tokens — real savings are 87% on short prompts and 35% on long ones.
- New effort setting, 1M-token context and big benchmark gains.
AI and ad platform news every day in our Telegram channel.
Sources: Anthropic, Claude API pricing, Claude Haiku 5.5 docs, 9to5Mac.