Anthropic Launches Low-Cost AI Model 'Claude Haiku 5.5'
TECHWORLD ·
✦ AI Summary
Anthropic launched the small AI model Claude Haiku 5.5 on the 7th.
The model is used for high-throughput tasks such as summarization, information compression, database queries, and classification, as well as real-time customer service and browser control.
The price is USD 0.1 per 1 million input tokens and USD 0.5 per 1 million output tokens, while requests exceeding 100,000 tokens are charged USD 0.5 per 1 million input tokens and USD 2.5 per 1 million output tokens.
Anthropic said on the 7th, local time, that it has launched the small AI model "Claude Haiku 5.5." "Claude Haiku 5.5" is designed to handle large-scale repetitive work at low cost. Anthropic described the model as the fastest among its released models.
The model is aimed at high-throughput tasks such as summarization, information compression, database queries, and classification. Its use cases include response-speed-sensitive work such as real-time customer service and browser control. In coding tasks, it serves as a sub-agent supporting Opus 5.5 and Sonnet 5.5.
The operating model is one in which higher-level models handle complex tasks, while Haiku handles repetitive or narrowly scoped work. This approach is designed to reduce the overall operating costs and response times of agents.
The pricing standard applies to input lengths of up to 100,000 tokens. The fee is USD 0.1 per 1 million input tokens and USD 0.5 per 1 million output tokens. Compared with Haiku 4.5, the input rate has been cut by 90%, and the output rate has also been cut by 90%.
Anthropic said that for requests exceeding 100,000 tokens, Haiku 5.5 will be charged at USD 0.5 per 1 million input tokens and USD 2.5 per 1 million output tokens. It also said that, when actual usage is reflected, average operating costs fall by about 75% compared with Haiku 4.5.
For the first time in the Haiku line, an effort setting for adjusting reasoning intensity has been introduced. Accordingly, users can adjust the setting to favor either cost reduction or higher-performance use, depending on the nature of the task.
In performance evaluations, scores improved from the previous model in knowledge work, computer control, and agentic coding. In Anthropic's own evaluation, GDPval-AA v2.1, it scored 1,620 points; in OSWorld 2.1, 72.4%; and in Terminal-Bench 4.0, 39.2%.
However, Anthropic said Sonnet 5.5 and Opus 5.5 are better suited for complex agentic coding. Haiku 5.5 is focused on narrow, repetitive tasks such as summarization, information compression, and sub-agents.
Anthropic said it lowered the cost burden by reducing the cost of using higher-tier models through price adjustments. The read price for Sonnet 5.5 cache has been cut in half, from USD 0.2 per 1 million tokens to USD 0.1 per 1 million tokens, and the company said the cost of typical agent tasks has fallen by about 20%.
Anthropic also said that, on the safety side, Claude Haiku 5.5 reduces instances of inappropriate behavior and responses to abuse requests compared with Haiku 4.5. It also said that in cybersecurity, it is broadening the range of defensive uses while applying restrictions to requests with a high likelihood of offensive use, such as penetration testing.
Claude Haiku 5.5 is available on the Claude platform, Amazon Web Services (AWS), Google Cloud, and Microsoft Azure. Anthropic said it plans to add computer and browser control features in beta to the Python SDK for developers and the TypeScript SDK for developers in order to expand the range of agent use.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=407930
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.