Anthropic has released an upgraded version of its most advanced artificial intelligence (AI) model, unveiling Claude Opus 4.6 with improvements aimed at complex coding, long-running tasks and high-value knowledge work.

In a recent blog post, the company said the new model builds on its predecessor with stronger software engineering capabilities, including more reliable performance in large codebases, improved code review and debugging, and the ability to sustain agent-like tasks for longer periods. For the first time in Anthropic’s Opus line, the model is being offered with a one-million-token context window, currently available in beta.

Anthropic also said that beyond programming, Opus 4.6 is designed for a broad range of professional tasks, such as financial analysis, research and working across documents, spreadsheets and presentations. Within Cowork, Anthropic’s environment for autonomous multitasking, the model can combine these functions to complete extended workflows on a user’s behalf.

According to the company, Opus 4.6 leads the Terminal-Bench 2.0 agentic coding evaluation and tops Humanity’s Last Exam, a multidisciplinary reasoning test. On GDPval-AA, which measures economically valuable knowledge work in areas such as finance and law, Anthropic said Opus 4.6 outperforms the next closest rival by a wide margin and also exceeds the performance of its own previous version. The model also ranks highest on BrowseComp, a benchmark focused on locating hard-to-find information online.

Claude Opus 4.6 is available through Anthropic’s API under the same pricing as before, at $5 and $25 per million tokens, depending on usage.