Claude Opus 4.6

Claude Opus 4.6 is Anthropic's flagship reasoning model built for complex, multi-step problem solving where accuracy outweighs the need for speed. Designed to sustain agentic workflows over longer horizons, it combines vision, function calling, and structured output capabilities to handle nuanced tasks ranging from coding and financial analysis to research and document creation. The model can review its own debugging output and catch mistakes—a feature that matters especially in large codebases where reliability compounds across files.

The model leads all frontier models on Humanity's Last Exam, a complex multidisciplinary reasoning test, and achieves the highest score on Terminal-Bench 2.0, the agentic coding evaluation. On GDPval-AA—an evaluation of economically valuable knowledge work in finance, legal, and related domains—Opus 4.6 outperforms OpenAI's GPT-5.2 by roughly 144 Elo points and its own predecessor by 190 points. It also outperforms every other model on BrowseComp, which tests a model's ability to locate hard-to-find information. These benchmark results position Opus 4.6 as a strong choice for knowledge-intensive work that demands sustained accuracy across long contexts.