Claude Haiku 5.5: Anthropic Cuts AI Prices by Up to 90%
Anthropic's new small model costs up to 90% less than its predecessor. Here's what Claude Haiku 5.5 can do, what it costs, and why cheaper AI matters.
AI is getting cheaper — fast. On October 7, Anthropic released Claude Haiku 5.5, the newest version of its smallest and fastest model, and dropped the price dramatically at the same time. For developers and businesses that run AI at scale, this is one of the biggest price moves of the year.
The new prices
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|
| Haiku 4.5 | $1.00 | $5.00 |
| Haiku 5.5 — prompts up to 100K tokens | $0.10 | $0.50 |
| Haiku 5.5 — prompts over 100K tokens | $0.50 | $2.50 |
That's a 90% cut for normal-length prompts and a 50% cut for very long ones. Anthropic says about 90% of requests to Haiku 4.5 were short enough to get the lower price.
There's one catch: Haiku 5.5 uses a newer tokenizer that counts the same text as roughly 30% more tokens. Taking that into account, Anthropic says the model costs around 75% less to run on average. Prompt caching (up to 90% savings) and batch processing (50% savings) can cut costs further.
What's new besides the price
- 1-million-token context window and up to 128,000 tokens of output.
- Adjustable effort: the first Haiku model that lets you trade speed and cost for more careful thinking. The default is medium.
- Better security: according to Anthropic's system card, it's substantially more resistant to prompt-injection attacks than Haiku 4.5 and refuses more malicious requests.
What it's good for
Anthropic positions Haiku 5.5 for high-volume, everyday work: summaries, classification, database queries, live customer support and browser automation. It's also designed to work as a sub-agent — a bigger model plans a job and hands smaller tasks to Haiku. For complex coding agents, Anthropic still recommends its larger Sonnet 5.5 and Opus 5.5 models.
Where you can use it
- The Claude API (model ID
claude-haiku-5-5), Amazon Bedrock, Google Cloud and Microsoft Foundry - Claude.ai on all plans, from Free to Enterprise
- Claude Code
More price changes in the same launch
Anthropic also halved the cache-read price of Sonnet 5.5 from $0.20 to $0.10 per million tokens, and added monthly API credits for paid subscribers: $100 for Max 5x, $200 for Max 20x, and up to $500 pooled across Team plans.
A note for developers
Upgrading isn't completely drop-in. Manual extended thinking with budget_tokens, non-default temperature, top_p and top_k values, and assistant message prefill now return errors or are no longer supported. Check Anthropic's migration notes before switching.
Why this matters
Haiku 5.5's price matches OpenAI's GPT-6 Luna, according to reports, so the AI price war is clearly heating up. For everyone else, falling prices mean AI features become cheaper to build into apps — which usually means more of them, in more places.
Are you using AI models in your own projects? Would a 90% price cut change what you build?
Sources: Anthropic, Claude Platform Docs, The New Stack, Let's Data Science. Image: Anthropic.
