Anthropic drops Haiku 5.5 at 90% off, somehow the bill still arrives on time 💸

Anthropic released Claude Haiku 5.5 on Wednesday, calling it the cheapest, fastest and most capable small model it has made, targeting high-volume tasks such as document summarization, database querying, live customer support and browser operation. Developers integrating Claude into their own products pay $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, the same rate OpenAI set for GPT-6 Luna when that rival small model launched on September 22. Haiku 4.5 had charged $1 and $5 per million tokens respectively, making the new pricing 90% lower, while prompts above 100,000 tokens receive a 50% cut. Because roughly 90% of requests to the prior model fell under that 100,000-token threshold, and because Haiku 5.5 splits text into slightly more tokens, Anthropic estimates average savings near 75%.

On benchmarks, Haiku 5.5 scored 72.4% on OSWorld 2.1, a partial-credit test of an AI's ability to operate a real computer through long, multi-step tasks, compared with 48.9% for OpenAI's GPT Luna. On Terminal-Bench 4.0, which scores AI agents on the share of professional tasks completed correctly via command-line on the first attempt, Haiku 5.5 reached 39.2%, against 16.4% for GPT Luna and 0% for Haiku 4.5, while Anthropic's Claude Sonnet 5.5 scored 70.6%. On GDPval-AA v2.1, an Elo-scale rating of real professional work across 44 occupations, Haiku 5.5 reached 1620, ahead of GPT Luna at 1437 and Haiku 4.5 at 735. Haiku 5.5 is also the first Haiku model to include an adjustable effort setting that lets users trade cost for smarter answers.

Anthropic simultaneously halved Sonnet 5.5's cache-read price to $0.10 per million tokens, the discounted rate applied to text a model has already processed. Haiku 5.5 arrives 15 days after Opus 5.5 launched on September 22 and nine days after Sonnet 5.5 launched on September 28, completing the three-model Claude 5.5 lineup Anthropic had promised. Opus 5.5 was the first release since CEO Dario Amodei published an essay urging the industry to slow gains in AI capabilities.

The new model is the first Haiku to ship with built-in safeguards covering a narrow range of high-risk cybersecurity requests, with Anthropic stating the controls target requests that could support harmful cyber activity while leaving most ordinary business and consumer tasks unaffected. Haiku 5.5 is available now on the Claude website, Amazon Web Services, Google Cloud and Microsoft Azure under the name claude-haiku-5-5. Anthropic is also rolling out monthly API credits this week: $100 for subscribers on the Max 5x plan and $200 for Max 20x. The launch comes ahead of Anthropic's expected November initial public offering at a near $2 trillion valuation.