Anthropic launches low-cost AI models! Sonnet 5.5 costs only half of Opus 5.5, a major upgrade in cybersecurity capabilities

The Zhitong Finance App learned that Anthropic released a new artificial intelligence (AI) model called Sonnet 5.5 on Monday. This is the second model the company has launched since CEO Dario Amoudi called for a slowdown in AI development.

According to reports, Sonnet 5.5 is a faster and lower cost product launched by Anthropic. The model went live on all platforms on Monday, including Amazon Web Services (Amazon Web Services), Google (GOOGL.US) Google Cloud, and Microsoft (MSFT.US) Azure. The price of Sonnet 5.5 is $2 per million input tokens and $10 per million output tokens, which is half the price of Opus 5.5. Anthropic said that compared to the previous generation version, Sonnet 5.5 also requires fewer tokens to complete each task, so running costs are lower.

Anthropic said that compared to the previous generation, Sonnet 5.5 performed better at programming, completing well-defined tasks, and producing beautiful documents, slides, and spreadsheets. This launch is less than a week since Anthropic launched the more expensive Opus 5.5. Anthropic also said on Monday that Haiku 5.5, the lowest-priced product line, is coming soon.

Theo Chu, product manager at Anthropic Research, said in an interview: “Sonnet really targets cost-conscious customers who probably don't need that high level of intelligence. Some tasks may just require performing routine work and don't require the kind of judgment that Opus can provide.”

Over the past few weeks, the safety and security measures of Anthropic and its rivals have come under close scrutiny as more researchers warn that artificial intelligence could cause catastrophic harm. Anthropic CEO Dario Amoudi called for a slowing down of cutting-edge AI model development in a cautionary post published on September 12. In the article, he said bluntly that the risks posed by AI are “serious” and that time must be taken to deal with these risks. The appeal was quickly answered by Musk and Altman, two key figures.

Anthropic said Monday that Sonnet 5.5 isn't driving the cutting edge of its model's capabilities, so according to a company blog post, most of the model's alignment tests have focused on “a specific set of risks applicable to any competency level model.” In the AI industry, an “aligned” (aligned) model refers to a model whose behavior is in line with human interests and values.

However, Anthropic said Sonnet 5.5's cybersecurity capabilities were “greatly improved” compared to previous generation products, so it was the first Sonnet model to use backup mechanisms and cybersecurity protection measures similar to those developed by Anthropic for its most powerful model at launch. “From the beginning, focusing on alignment and safety was a critical part of our mission,” said Theo Chu. This has always been a priority for us in all of our models.”