Anthropic has rolled out Claude Haiku 5.5, refreshing the smallest and most economical tier in its language model family. The update lands roughly twelve months after the previous Haiku release and marks Anthropic's third 5.5 model upgrade in the span of a month, following recent revisions to Opus and Sonnet.
The headline shift here is cost. Anthropic claims Claude Haiku 5.5 costs around 75% less to run than the Claude Haiku 4.5 model it replaces. That drop targets developers running automated systems where query volume piles up quickly.
Anthropic built Haiku specifically for cost-sensitive, high-throughput jobs. The company points to repetitive tasks like database queries, content compactions, data classifications, and quick text summaries as ideal fits. Because Haiku 5.5 is also pitched as Anthropic's fastest model so far, the company is positioning it for speed-critical workflows such as browser automation and live customer support bots. It is also designed to is a secondary subagent handling background coding duties alongside the larger Opus 5.5 and Sonnet 5.5.
Anthropic structures its core lineup into three distinct sizes. Haiku sits at the bottom as the leanest option. Sonnet occupies the middle tier for balanced everyday performance, while Opus sits at the top among the primary three. Sitting above all of them is Anthropic's Fable and Mythos tier, designed for the most demanding enterprise compute demands.
Cheaper Sonnet Caching and New API Credits
Alongside the Haiku 5.5 rollout, Anthropic is tweaking pricing structures across the rest of its developer catalogue. The company has halved the cost of cache reads on Claude Sonnet 5.5. According to Anthropic, that specific price cut brings running expenses down by around 20% across most agentic workflows.
Subscribers on Claude Max and Claude Team tiers are also getting a new monthly allowance of API credits. Anthropic says those credits are intended to encourage developers to build and test new autonomous agents directly on its platform.
The update gives developers immediate access to cheaper compute across both small-scale automations and multi-tier agent pipelines.
ADFiled by The AI Desk
Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.
More from this desk →
Be the first to comment
Join the argument. No password, just your email or a passkey.