Anthropic has introduced an enhanced iteration of its cutting-edge artificial intelligence (AI) model, unveiling Claude Opus 4.6 with enhancements targeting intricate coding, prolonged tasks, and high-value knowledge-intensive assignments.
In a recent blog update, the corporation highlighted that the new model builds upon its forerunner by bolstering its software engineering capabilities. This includes heightened performance reliability within extensive codebases, enhanced code assessment and debugging, and the capacity to sustain agent-like duties over extended durations. A significant feature of Anthropic’s Opus series is the introduction of a one-million-token context window, currently accessible in beta.
Anthropic also emphasized that beyond its application in programming, Opus 4.6 is tailored for a diverse array of professional responsibilities, such as financial analysis, research tasks, and collaboration across various documents, spreadsheets, and presentations. In Anthropic’s Cowork platform, designed for self-directed multitasking, the model can integrate these functions to execute comprehensive workflows autonomously on behalf of the user.
The company reported that Opus 4.6 excels in the Terminal-Bench 2.0 assessment for agentic coding and surpasses Humanity’s Last Exam, a comprehensive reasoning evaluation. On GDPval-AA, a metric assessing economically significant knowledge work in fields like finance and law, Anthropic claimed that Opus 4.6 outperforms its closest competitor by a considerable margin and also surpasses the performance of its previous iteration. Additionally, the model achieves the highest ranking on BrowseComp, a benchmark concentrating on retrieving obscure information online.
Claude Opus 4.6 can be accessed through Anthropic’s API at the same pricing structure as before, priced at $5 and $25 per million tokens based on the level of usage.
