Anthropic has launched its new AI model, Claude Opus 4.6, as a major update to its advanced model series. This release addresses many of the limitations that existed in previous versions, primarily improving performance in handling complex tasks and precise technical requirements. The new model is an evolution of the basic Opus 4 version, with a clear focus on enhancing its capabilities in the areas of software engineering and handling long contexts and large data sets.
The previous version, Claude Opus 4.5, had good advanced thinking capabilities, but faced challenges when processing tasks that required handling extended information or large databases. The new version addresses this issue by providing an expanded context window of up to one million symbols, a significant improvement over the previous limit of 200,000 symbols, allowing the model to process vast amounts of data without degradation in performance.
The company announced that Claude Opus 4.6 includes a set of new technical features, including a context compression feature that allows old data to be summarized and intelligently updated during long interactions, as well as an adaptive reasoning feature that enables the model to assess the complexity of a question and allocate an appropriate amount of intellectual processing to it. The company has also added options to control the level of effort from low to maximum, striking a balance between speed, accuracy, and cost.
In terms of performance, Anthropic confirmed that the new model achieved advanced results in several internal tests and benchmarks, topping the Terminal-Bench 2.0 tests for command line skills, as well as Humanity's Last Exam tests for multidisciplinary thinking. In interactive evaluations, Claude Opus 4.6 outperformed competing models such as GPT-5.2 by a significant margin in evaluation scores, and showed significant improvement over the previous version, Opus 4.5, especially in the financial and legal domains.
In programming tests such as SWE-bench Verified, the model scored an average performance of over 81 percent, reflecting its high ability to independently handle software development tasks, debugging, and managing large software repositories. CyberGym tests also showed strong performance in the fundamental aspects of cybersecurity, even without the need for advanced levels of reasoning.
The company emphasized that security was a key element in the development of the new model, with Claude Opus 4.6 undergoing extensive security reviews that showed low rates of errors and undesirable behaviors such as deception or overreaction. Six new cybersecurity-specific tests have also been added to monitor for potential misuse and enhance the model's ability to support defensive applications and detect vulnerabilities in open-source code.
In practical applications, the model has advanced capabilities in managing complex software projects, conducting thorough code reviews, and creating virtual development teams that work in parallel across the Claude Code platform. In business environments, it can perform advanced financial analysis, prepare professional reports and documents, and handle multi-step searches using tools such as Claude within Excel, with improved support for unstructured data and long-term tasks.
The company has also expanded the model's capabilities to include working on presentations via a research version of Claude within PowerPoint, as well as advanced uses in areas such as computational biology and scientific research, where the new version has shown nearly twice the performance of Opus 4.5, opening the door to more complex and accurate scientific applications.
Anthropic announced that Claude Opus 4.6 is now available to users via the company's website, through mobile and computer applications, as well as through the API and major cloud service providers. In terms of pricing, the cost of using the API is set at $5 per million input tokens and $25 per million output tokens, with additional fees for using extended contexts.
The launch of this model represents a new step in the race to develop advanced artificial intelligence systems, reflecting the growing trend toward building models that are more intelligent and capable of understanding complex contexts, while maintaining the highest standards of security and reliability, enabling their widespread use in business, programming, and scientific research.