Upgrading our smartest model.
The new Claude Opus 4.6 significantly improves the coding capabilities of previous models. It plans more carefully, sustains agent tasks for longer periods, operates more reliably on large codebases, and has enhanced code review and debugging abilities to catch its own mistakes. Additionally, for the first time in an Opus-class model, Opus 4.6 offers a 1 million token context window in beta.
Opus 4.6's improved capabilities can be applied to a variety of everyday tasks: performing financial analysis, research, utilizing and generating documents, spreadsheets, and presentations. In Cowork, where Claude can autonomously multitask, Opus 4.6 can leverage all these capabilities on behalf of the user.
This model's performance is at the cutting edge in several benchmarks. For example, it achieved the highest score on Terminal-Bench 2.0, an agent coding benchmark, and leads all frontier models on Humanity's Last Exam, a complex multidisciplinary reasoning test. In GDPval-AA, which measures the performance of economically valuable knowledge work tasks such as finance and law, it outperformed the industry leading model (OpenAI's GPT-5.2) by approximately 144 Elo points and its previous iteration (Claude Opus 4.5) by 190 points. Opus 4.6 also demonstrates superior performance compared to any other model on BrowseComp, which measures a model's ability to search for rare information online.
<< Truncated >>
Translated with Claude Opus 4.6.
▶ Original source: https://www.anthropic.com/news/claude-opus-4-6
