Anthropic releases Claude Opus 4.6 with million-token context and enhanced reasoning

Anthropic has unveiled Claude Opus 4.6, the latest version of its most advanced AI model, featuring a 1 million token context window for the first time in the Opus lineup. The update brings significant improvements in coding capabilities, long-context information retrieval, and autonomous task handling. According to the company, the model plans more carefully, sustains agentic tasks for longer periods, and operates more reliably in larger codebases with enhanced code review and debugging abilities.

Performance benchmarks show Claude Opus 4.6 leading on several industry evaluations. The model achieved the highest score on Terminal-Bench 2.0, an agentic coding test, and topped all frontier models on Humanity's Last Exam, a complex multidisciplinary reasoning assessment. On the MRCR v2 needle-in-a-haystack benchmark testing long-context retrieval, Opus 4.6 scored 76% compared to Sonnet 4.5's 18.5%, representing what Anthropic describes as "a qualitative shift in how much context a model can actually use while maintaining peak performance." The model also outperformed its predecessor by 190 Elo points on GDPval-AA, which measures performance on economically valuable knowledge work across finance, legal, and other professional domains.

The release includes new API features designed for developer flexibility. Adaptive thinking allows the model to automatically determine when deeper reasoning is needed, replacing the previous binary on-off choice for extended thinking. Four effort levels — low, medium, high, and max — give developers control over computational intensity, with the company recommending users dial down from the default high setting if the model overthinks simpler tasks. Context compaction in beta automatically summarizes older context when conversations approach size limits, enabling longer-running tasks without hitting constraints. 

Claude Opus 4.6 also expands into productivity tools with substantial upgrades to Claude in Excel and the introduction of Claude in PowerPoint as a research preview. The Excel integration now handles long-running tasks with improved planning capabilities, processes unstructured data without guidance, and executes multi-step changes in single passes. The PowerPoint integration reads layouts, fonts, and slide masters to maintain brand consistency when building presentations from templates or generating complete decks from descriptions. In Claude Code, developers can now assemble agent teams that work in parallel and coordinate autonomously on tasks like codebase reviews.

For everyday users, these updates address common frustrations with AI assistants losing track of information or failing to complete complex tasks. The expanded context window means the model can handle much longer conversations and larger documents without forgetting earlier details, while improved long-context retrieval helps it find specific information buried in extensive materials. Safety evaluations show Opus 4.6 maintains low rates of problematic behaviors while reducing instances where the model unnecessarily refuses to answer legitimate questions. 

Try Closer AI website builder, with the most advanced AI technologies built in.