Ten cents per million input tokens and 50 cents per million output. That is the new price of Anthropic’s smallest model, Claude Haiku 5.5, released on Wednesday. Its predecessor cost $1 and $5.
The 90 percent cut comes with a condition. VentureBeat reported that it applies to prompts of up to 100,000 tokens. Above that, prices rise to 50 cents and $2.50, which is half the old rate.
Anthropic says about 90 percent of requests to the previous Haiku already fell under the threshold. The lower tier matches what OpenAI charges for GPT-6 Luna.
The company also halved the cost of cached reads on its mid-range Sonnet 5.5 model. Haiku 5.5 is available through Anthropic, AWS, Google Cloud and Microsoft Azure.
On Anthropic’s own benchmarks the model is a large step. It scored 39.2 percent on Terminal-Bench 4.0, an agentic coding test on which Haiku 4.5 scored zero.
Small models are where volume sits. They run the routine steps inside AI agents, which call a model thousands of times per task, so price per token decides which products are viable. This is the second round of cuts in three weeks, after OpenAI and Anthropic repriced their flagship models on the same day in September.
The asterisk is in Anthropic’s own estimate. It expects typical workloads to cost about 75 percent less, not 90, once request sizes and changes in token consumption are counted. Neowin reported the new model uses far more tokens to do the same work. How many more was not stated, and the benchmark scores have yet to be reproduced by anyone outside the company.
