Anthropic Releases Claude Opus 4.7, Better at Coding but "Restrained" for Safety

Anthropic, an artificial intelligence company, announced its latest AI model, Claude Opus 4.7, on Thursday (16 April 2026). Claude Opus 4.7 is the successor to Claude Opus 4.6, which was rolled out in February 2026. According to Anthropic, compared to its predecessor, Claude Opus 4.7 offers better performance in software engineering, following instructions, and completing real-world tasks. “Opus 4.7 handles complex and long-duration tasks meticulously and consistently, paying attention to instructions quickly and performing its own output verification before reporting,” stated Anthropic, as quoted by KompasTekno from its official website. In addition, this model is also claimed to be more sophisticated and creative when completing professional tasks, creating interfaces, slides, and documents. Furthermore, Opus 4.7 is claimed to excel in various usage scenarios, including industry benchmarks for agentic coding, multi-disciplinary reasoning, large-scale tool usage, and agent-based computer interactions. Opus 4.7 also achieved 87.6% on SWE-bench Verified compared to Opus 4.6. In terms of multi-disciplinary reasoning measured on the Humanity’s Last Exam (HLE) benchmark, Opus 4.7 scored up to 54.7%, outperforming Opus 4.6 (53.3%) and Gemini 3.1 Pro (51.4%). However, the model’s cybersecurity capabilities are designed not to be as advanced as Anthropic’s experimental model, Claude Mythos Preview. “We are releasing Opus 4.7 with safeguards that automatically detect and block requests related to prohibited or high-risk cybersecurity uses,” explained Anthropic. The company added that experiences from implementing these safeguard systems in the real world will form the basis for further development before it releases models with cybersecurity capabilities equivalent to Mythos more widely. Claude Opus 4.7 is now available across all Claude products from Anthropic, including via API, as well as on cloud platforms from Microsoft, Google, and Amazon. Its price is the same as Opus 4.6, namely 5 US dollars (approximately Rp 85,000) per one million input tokens and 25 US dollars (approximately Rp 428,000) per one million output tokens.