Anthropic has officially launched Claude Haiku 5.5, the latest lightweight model in the Claude 5.5 series. For regular requests within 100,000 Tokens, its input and output prices drop to $0.10 and $0.50 per million Tokens respectively, a substantial 90% decrease from the previous generation's list price.
As the currently fastest-responding model, Haiku 5.5 targets high-concurrency, low-latency, and high cost-effectiveness scenarios, making it well suited for high-frequency tasks such as real-time customer service, summarization and extraction, and data classification. In addition, the system for the first time introduces adaptive thinking and dynamic adjustment functions, able to automatically balance reasoning depth and computing cost according to task complexity.
New Tokenization Mechanism Draws Attention; Actual Consumption Costs May Increase
However, the company also cautions that because Haiku 5.5 uses a brand-new tokenizer shared with the advanced models, the number of Tokens converted from the same text input has risen. Tests show that long-text tasks may consume 25% to 30% more Tokens, so after accounting for actual workloads, the overall operating cost reduction is approximately 75%.
Note that once a single request exceeds the 100,000 Token threshold, both its input and output unit prices soar to 5 times the base price. This means that while developers enjoy the low unit-price advantage, they must also precisely evaluate Token consumption for specific workflows to avoid falling into the hidden cost trap.