Anthropic: Claude Opus 5's ability to detect cyber-security holes close to the top model Mythos 5

On 25 July, Anthropic released a new generation of artificial intelligence model Claude Opus5, stating that the model ' s ability to detect network security gaps was almost comparable to the company ' s more advanced Mythos5 model. Anthropic states that Claude Opus 5 is a generic model that has been upgraded in such tasks as self-censorship, scientific research and code optimization. According to the company, although the model is close to Mythos5 in its ability to detect loopholes, it is still "manifestly behind" in the actual use of these loopholes. In addition, Anthropic emphasizes that Claude Opus 5 performed in many capacity categories close to its most advanced Fable5 model, but at a cost of only about half the latter, and is expected to be the default option for an enterprise to handle day-to-day office tasks. As businesses continue to focus on AI input costs, several AI companies, including OpenAI, Meta and SpaceX AI, are accelerating competition for more cost-effective models. Anthropic states that the market's focus on AVP and practical application is increasing。