핵심 요약
- 1. Anthropic introduced Claude Opus 5, which demonstrates performance close to the Fable 5 level.
- 2. The new model's performance in some programming tests has doubled compared to its predecessor Opus 4.8.
- 3. Despite the performance increase, the usage cost did not rise; in specific tests, costs decreased.
- 4. To address security concerns, the model is equipped with stronger protection measures.
숫자로 보기
Anthropic has released the new artificial intelligence model Claude Opus 5. The company states that the new model approaches the performance of Fable 5, but its usage cost is not as high. It also includes significant improvements in areas such as software engineering and rational reasoning efficiency.
Compared to the previous version, Opus 4.8, Opus 5 doubled its performance in some programming tests. Despite this, the usage price of the model did not increase; in fact, the cost per task in the Frontier-Bench v0.1 test is seen to be lower than its predecessor. This situation makes the new model highly attractive for developers.
The artificial intelligence tool also yields better results in tasks such as multidisciplinary reasoning, search, and computer use. Anthropic added stronger security measures to the new model than its predecessors to prevent the misuse of advanced capabilities. The company emphasizes that the model is more resistant to being deceived for harmful purposes.
반응 남기기
이 기사에 대해 질문
답변은 이 기사만을 바탕으로 AI가 생성합니다.
자주 묻는 질문
- What are the prominent performance improvements of Claude Opus 5?
- The new model provides significant performance increases, especially in areas such as programming tasks, multidisciplinary reasoning, and scientific research (primarily biology).
- How is the usage cost of Opus 5 compared to previous models?
- Although the model approaches the Fable 5 level, it is more cost-effective. Additionally, in specific tests, the cost per task is lower compared to Opus 4.8.
- What measures did Anthropic take regarding security?
- To prevent the misuse of advanced capabilities, stronger security measures than in previous versions were added to the model, and the system's deceptibility was reduced.
다른 출처의 보도 · 1
- Memeburn·