Anthropic Launches Claude Opus 4.6

Anthropic on Thursday released Claude Opus 4.6, a significant upgrade to its Opus foundation model lineup. The new artificial intelligence (AI) model is designed to overcome limitations seen in earlier versions, particularly in handling long and complex tasks across software engineering, finance, law, and other knowledge intensive domains.

A key highlight of Claude Opus 4.6 is its beta support for a one million token context window, a major jump from the 200,000 token limit in previous Opus models. This enables the model to process vast datasets while maintaining stable performance during extended interactions. New features such as context compaction help summarise and refresh older information, while adaptive thinking allows the model to adjust reasoning depth based on task complexity.

On internal benchmarks, Anthropic says Claude Opus 4.6 delivers industry-leading results. The model tops Terminal Bench 2.0 for command-line tasks and Humanity’s Last Exam for multidisciplinary reasoning. In agent-based evaluations, it reportedly outperforms OpenAI’s GPT 5.2 and its predecessor Opus 4.5, with strong results in finance, legal reasoning, cybersecurity tests, and verified software engineering benchmarks.

Anthropic emphasised that safety remains a priority, with the model showing low rates of deception and over-refusal, alongside new cybersecurity safeguards. Claude Opus 4.6 is capable of managing large codebases, conducting reviews, debugging, and supporting business workflows such as financial analysis and document generation. The model is now available via Anthropic’s apps, API, and major cloud platforms, with pricing starting at $5 per million input tokens and $25 per million output tokens.