Anthropic made its AI model stronger than ChatGPT's
Anthropic has launched its flagship model Claude Opus 5.5, which outperforms OpenAI's GPT-6 Astra in code writing and complex tasks at a cost tens of percent lower. The model includes advanced defense mechanisms that reduce escape attempts from experimental environments by 85%. The launch marks another step in the AI race, alongside calls from the company's CEO for a slowdown.
Anthropic launched yesterday (Tuesday) Claude Opus 5.5, the first flagship model in the 5.5 series, showing a significant leap in programming, data analysis, and computer control capabilities. In official performance tests, the model surpasses OpenAI's GPT-6 Astra in autonomous code writing (Terminal-Bench 4.0), graph understanding, and complex office tasks (GDPval-AA). The model produces output 30% faster than its predecessor, and its operation is 40% cheaper than Opus 5 and 60% cheaper compared to OpenAI's equivalent model. In early test experiments, the system performed a full migration of 680,000 lines of code in less than a day and reduced loading times on 39 out of 40 web pages. The launch directly addresses industry security failures: the model was tested by METR and Frontier Design and recorded an 85% decrease in attempts to bypass experimental environment boundaries. Anthropic integrated a "protected thinking" mechanism that thwarts distillation attacks from competitors, alongside a zero data retention policy and digital watermarking. The model is available starting today on the Claude website, AWS Bedrock, Google Cloud, and Microsoft. The rapid launch proves that the race for technological supremacy continues to set the pace, even as industry leaders warn of AI dangers.