Anthropic launched Claude Haiku 5.5 on Tuesday, completing its current model lineup with a small model built for high-volume, cost-sensitive workloads. The launch comes ahead of the company’s planned IPO and follows the earlier releases of Sonnet 5.5 and Opus 5.5.
With Haiku 5.5, it is the pricing which is grabbing eyeballs. For requests under 100,000 tokens — which account for roughly 90% of requests to the previous Haiku model — Haiku 5.5 costs 90% less than Haiku 4.5. Input tokens are priced at $0.10 per million, output at $0.50 per million, and cache reads at $0.01 per million. At the previous model’s pricing, those same numbers were $1.00, $5.00, and $0.10 respectively. Anthropic says the average cost reduction across typical usage works out to around 75%.
On benchmarks, Haiku 5.5 scores 72.4% on OSWorld 2.1 — a test of how well AI agents complete real computer tasks — compared to 15.7% for Haiku 4.5 and 48.9% for OpenAI’s GPT-6 Luna. On Humanity’s Last Exam, a test of expert-level reasoning, Haiku 5.5 reaches 45.9% without tools and 57.4% with tools, against Haiku 4.5’s 10.2% and 18.7%. On agentic coding via Terminal-Bench 4.0, Haiku 5.5 scores 39.2% — the previous generation scored zero.
The practical positioning is clear from how Anthropic and its early customers are describing it. Haiku 5.5 is not meant to replace Sonnet or Opus for complex reasoning tasks. It is designed to sit below them in agentic pipelines — handling subagent work, summarisation, compaction, classification, and quick lookups while a larger model handles the strategic layer. Cognition, which makes the AI coding agent Devin, confirmed Haiku 5.5 as a subagent option in Devin Fusion, with the combination delivering a FrontierCode score of 66.2 at reduced cost and latency. Asana reported over 30% reduction in latency and up to 2.5x faster inference per agent turn. HubSpot put it through CRM tasks and recorded a 92.8% accuracy score — the best it had seen on that suite.
Haiku 5.5 is also Anthropic’s first Haiku-class model to support an adjustable effort setting, allowing developers to trade off cost against intelligence depending on the task. That feature had previously been limited to the larger models.
Alongside the Haiku launch, Anthropic cut the price of cache reads on Sonnet 5.5 by 50% — from $0.20 to $0.10 per million tokens. Because cache reads make up a significant share of token consumption in agentic work, Anthropic says this reduces the effective cost of Sonnet 5.5 on most agentic tasks by around 20%.
The company also announced a new monthly API credit for Max and Team subscribers. Max 5x users get $100 in monthly credits, Max 20x users get $200, and Team subscribers receive up to $500 pooled across users. The credits can be used on any Claude model and are designed for building tools, apps, and agents on the Claude Platform.
Claude Haiku 5.5 is available now across Anthropic’s API and on AWS, Google Cloud, and Microsoft Azure.