Claude Haiku 5.5: Pricing cuts and real costs

According to Anthropic, the new Claude Haiku 5.5 model reduces input and output token costs by up to 90 percent for prompts under 100,000 tokens. For development teams running high-volume customer support bots or automated database queries, this price adjustment directly lowers monthly API overhead.

How much do token costs actually drop?

While per-token prices fall significantly, the model introduces an updated tokenizer. Similar to previous architecture updates where token consumption rose by about 30 percent, this new tokenizer requires more tokens to process the same text. Real-world savings will therefore sit below the headline 90 percent reduction.

What changes for API workflows on Monday morning?

Development teams can now adjust reasoning levels within the Haiku class for the first time. This lets engineers balance API expenses against execution quality for narrowly scoped tasks like text summarization or sub-agent routing.

• Input tokens for short prompts drop to $0.10 per million tokens.
• Cache read costs for Sonnet 5.5 are cut in half to $0.10 per million tokens.
• Monthly API credits ranging from $100 to $500 are now available for subscriber tiers.