Claude Haiku 5.5 cuts API prices by up to 90%

Anthropic has released Claude Haiku 5.5, lowering the cost of using its small AI model by up to 90%.

Haiku is designed for quick, repetitive tasks such as summarizing documents, sorting information and answering customer questions. Anthropic says the new version is its fastest model at standard speed, though Opus in Fast Mode remains quicker.

Developers pay for API use based on how much information they send to the model and how much it generates. For prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens, compared with $1 and $5 for Haiku 4.5. Longer prompts cost $0.50 and $2.50 respectively, half the previous model’s rates, according to Anthropic’s pricing.

The company estimates that running Haiku 5.5 costs around 75% less on average. Actual savings depend on the workload, partly because the new model divides text into more tokens than its predecessor. Developers switching models therefore need to recalculate their usage.

Anthropic’s published tests also show improvements in the model’s ability to operate computers and complete coding tasks.

Computer use: Haiku 5.5 scored 72.4% on OSWorld 2.1’s offline subset, compared with 15.7% for Haiku 4.5 and 48.9% for OpenAI’s GPT-6 Luna.
Coding: It scored 39.2% on Terminal-Bench 4.0, which tests tasks performed through a command-line interface, compared with 0% for Haiku 4.5 and 16.4% for GPT-6 Luna.

The larger Sonnet 5.5 scored higher on both tests. Haiku 5.5 also lets developers adjust how much reasoning the model does, choosing between lower costs and more computation for difficult requests.

Anthropic has separately cut the cost of reusing information with Sonnet 5.5. Its cache-read price has fallen from $0.20 to $0.10 per million tokens, making it cheaper for applications to repeatedly use the same documents or instructions. Anthropic estimates this reduces costs by around 20% on most tasks performed by AI agents.

Max and Team subscribers are also getting monthly API credits to build their own apps and tools. Max 5x includes $100, Max 20x includes $200, and Team accounts receive a shared balance of up to $500. The credits do not increase message limits in the Claude app.

Haiku 5.5 is available through the Claude Platform, AWS, Google Cloud and Microsoft Azure.