Anthropic tung mô hình AI mới: Hiệu năng cao, chi phí rẻ hơn một nửa

VTV.vn - Anthropic launches a new AI model with performance approaching the high-end tier but at half the cost.

On July 24, the artificial intelligence (AI) startup Anthropic launched its new AI model, Claude Opus 5, which is rated as having performance nearing the company's most powerful system while costing 50% less. This move comes as businesses increasingly focus on cost optimization when deploying AI applications.

According to Anthropic, Claude Opus 5 achieves top-tier results in benchmark tests for programming and knowledge-based task processing, while consuming fewer computational resources compared to previous versions. The San Francisco-based (USA) company stated that the new model is optimized for daily office tasks and computer programming.

Speaking to the press, Dianne Penn, Director of Product at Anthropic, emphasized that this is the next development following the Opus 4.8 version released in May, reflecting the company's rapid pace of innovation. Ms. Penn stated: "We continue to develop advanced AI models and strive to bring this technology to users in the most accessible way possible with each product generation."

Regarding costs, Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, just 50% of the cost of Fable 5 - the company's current most expensive commercial model. Despite this, in many tasks, the performance of Opus 5 remains nearly equivalent to Fable 5.

The cost-reduction strategy is considered a notable move in the context of businesses becoming increasingly cautious with their AI investments. Currently, Anthropic and its top competitor, OpenAI, both filed for stock market listings in June, causing investors to closely monitor the business performance of both companies.

Beyond performance, Anthropic also emphasized safety. The company stated that Opus 5 was not extensively trained for cyberattack tasks and has significantly lower capabilities than the high-end Mythos 5 AI line in exploiting security vulnerabilities for cyberattacks or purposes that could be misused.

According to Anthropic, although Opus 5 has software vulnerability detection capabilities nearly equivalent to Mythos 5, the model is much less effective at turning those vulnerabilities into attack tools. Thanks to the lower risk level, Opus 5's automated safety filter system will intervene approximately 85% less than Fable 5. Previously, Fable 5 and Mythos 5 were temporarily suspended from public availability in June after the U.S. Department of Commerce imposed export controls on certain advanced AI models. These measures were subsequently lifted at the end of June.