Anthropic released Claude Haiku 5.5 on Wednesday, calling it the cheapest, fastest, and most capable small model in its lineup to date. The model is aimed at high-volume work such as summarizing documents and querying databases, as well as speed-sensitive tasks including live customer support and operating a web browser for a user.
Pricing drops sharply from Haiku 4.5
Developers integrating Claude into their own products will pay $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Tokens are the small units of text AI models process, and they are the basis for how major AI companies charge users. According to the report, that rate matches the pricing OpenAI used for GPT-6 Luna when it launched on Sept. 22.
Haiku 4.5 had been priced at $1 per million input tokens and $5 per million output tokens, so the new rates are 90% lower. Anthropic also cut pricing by 50% for prompts above 100,000 tokens. The company said about 90% of requests to the older model were under that threshold, and because Haiku 5.5 splits text into slightly more tokens, it estimated the average savings at close to 75%.
Built for repetitive support and summarization work
Anthropic said customer-support chats and long-email summaries are the kinds of jobs Haiku 5.5 was built for. In the report’s framing, repetitive, easy, and non-creative tasks are the model’s best fit.
Benchmark scores top OpenAI Luna
Anthropic also pointed to benchmark results showing the model is not limited to basic tasks. On OSWorld 2.1, which measures whether an AI can operate a real computer through long, multi-step tasks and reports performance as a partial-credit percentage, Haiku 5.5 posted a 72.4% success rate. OpenAI’s GPT Luna scored 48.9%.
On Terminal-Bench 4.0, where AI agents are given professional tasks to complete by entering commands on their own and are scored by the share completed correctly on the first attempt, Haiku 5.5 scored 39.2%. OpenAI Luna scored 16.4%, while Haiku 4.5 scored 0%. For comparison, Anthropic’s Claude Sonnet 5.5 scored 70.6%.
On GDPval-AA v2.1, which rates models on real professional work across 44 occupations using an Elo system borrowed from chess, Haiku 5.5 scored 1620. Luna scored 1437, and Haiku 4.5 scored 735.
First Haiku model with adjustable effort
This is also the first Haiku model to include an adjustable effort setting, which lets users trade cost for more capable answers. Anthropic also halved Sonnet 5.5’s cache-read price to $0.10 per million tokens, the discounted rate for text the model has already processed.
Release timeline and availability
Haiku 5.5 arrived 15 days after Opus 5.5, which launched on Sept. 22, and nine days after Sonnet 5.5, which launched on Sept. 28. The report said it is the last of the three Claude 5.5 models Anthropic had promised. It also noted that Opus 5.5 was the first release since CEO Dario Amodei published an essay urging the AI industry to slow gains in capabilities.
Haiku 5.5 is now available on the Claude website, Amazon Web Services, Google Cloud, and Microsoft Azure under the name claude-haiku-5-5.
Monthly API credits roll out this week
Anthropic is also rolling out monthly API credits this week. Subscribers on the Max 5x plan will receive $100, Max 20x users will receive $200, and Team plans will get up to $500 pooled across users.
Report test found fast responses but an incorrect answer
The report said it tried the model on a simple logic question and got a response almost instantly. The answer was wrong, however, and the article cautioned against blindly trusting the model’s output.