Anthropic has unveiled Claude Haiku 5.5; a model designed with a focus on high speed, low cost, and handling large volumes of requests, intended for tasks such as summarization, classification, information retrieval, database queries, and running agent-based AI tasks.
Anthropic says Haiku 5.5 is the fastest, cheapest, and most powerful Haiku model the company has released to date. The model was introduced on October 7, 2026 (15 Mehr 1405) and is the third member of the Claude 5.5 family after Opus 5.5 and Sonnet 5.5. According to Anthropic, the average running cost of Haiku 5.5 is about 75 percent lower than the previous generation, Haiku 4.5.
Focus on speed and high-throughput processing
One of the important features of Claude Haiku 5.5 is the addition of an adjustable effort level setting for the first time in the Haiku family. This capability allows developers to choose between higher speed and lower cost or using more compute for deeper reasoning.
This approach makes Haiku 5.5 suitable for fast, repetitive tasks such as summarization, text compression, classification, and database queries. At the same time, Anthropic recommends it for latency-sensitive applications, including live customer support and use within a computer and browser.
Anthropic introduces Haiku 5.5 as its standard, fastest model; although some of the company’s larger models can be faster in Fast Mode.
In the OSWorld 2.1 benchmark for evaluating computer interaction ability, Haiku 5.5 achieved a score of 72.4 percent; while Haiku 4.5 had scored 15.7 percent. In the Terminal-Bench 4.0 test, which evaluates models’ performance on agent-driven programming tasks, Haiku 5.5 scored 39.2 percent.
The model can also act as a sub-agent alongside larger models like Opus 5.5 and Sonnet 5.5; for example, the larger model can decompose a complex task into several smaller tasks and Haiku 5.5 can handle the simpler, high-volume portions.
Significant reduction in model running cost
The most important change in Haiku 5.5 is not only the speed increase; Anthropic has also tried to noticeably reduce its usage cost.
For requests up to 100,000 tokens, the cost is $0.10 per million input tokens and $0.50 per million output tokens. These figures in Haiku 4.5 were $1 and $5 respectively; therefore the cost for this category of requests has been reduced by 90 percent.
For requests over 100,000 tokens, the input cost is $0.50 and the output cost is $2.50 per million tokens, which is about 50 percent less compared to Haiku 4.5. Anthropic says about 90 percent of previous model requests fell under the 100,000-token range.
The company simultaneously reduced the Cache Read price in Claude Sonnet 5.5 from $0.20 to $0.10 per million tokens. According to Anthropic, this change can lower the running cost of Sonnet 5.5 in many agent-driven tasks by about 20 percent.
Access to Claude Haiku 5.5
Claude Haiku 5.5 is now available through the Claude platform, and developers can use it on services such as Amazon Web Services, Google Cloud, and Microsoft Foundry. The model is also available in Claude Code.
Anthropic has also allocated monthly API credits for Max and Team subscribers; Max 5x subscribers receive $100 monthly, Max 20x subscribers receive $200 monthly, and Team users receive up to $500 in total credit. These credits can be used for API usage and building AI tools, apps, and agents.
Anthropic’s Python and TypeScript packages have also been updated with experimental support for the computer and browser usage features.
Overall, Haiku 5.5’s goal is not to directly compete with larger models on the most complex tasks; rather, Anthropic aims to make running millions of small, repetitive tasks more economical by providing a fast, low-cost model. For this reason, Haiku 5.5 can sit alongside Sonnet and Opus in agent-based architectures and perform tasks for which using larger models would not be cost-effective.