Anthropic rolls out upgraded Claude Opus model with improved coding and agentic AI skills
Claude Opus 4.6 can plan more carefully and sustain agentic tasks for longer periods than its predecessor. It can also operate more reliably in larger codebases and has better code review and debugging skills.
Anthropic has released Claude Opus 4.6, an upgrade to its flagship AI model with improved coding skills and the ability to sustain agentic tasks.
The San Francisco-based AI startup said the new model can plan more carefully and sustain agentic tasks—wherein the AI operates autonomously to achieve a goal—for longer periods than its predecessor Opus 4.5. It can also operate more reliably in larger codebases and has better code review and debugging skills to catch its own mistakes, the company said.
Anthropic claims Opus 4.6 outperforms competitors on several industry benchmarks. On GDPval-AA, a test measuring performance in economically valuable tasks across finance and law, the company reported Opus 4.6 scored approximately 144 Elo points higher than the next-best model, OpenAI’s GPT-5.2.
The release comprises several new features aimed at enterprise and developer workflows, including Agent Teams, a research preview within Claude Code that allows users to deploy multiple AI agents that split tasks and coordinate autonomously.
PowerPoint integration enables AI to read layouts and slide masters to generate presentations, while the 1 million token context window, now in beta, allows the model to process significantly larger documents and datasets.
Anthropic has also introduced 'effort controls' that allow developers to choose from four settings—low, medium, high, and max—to optimise token usage with the depth of the model's reasoning. A new 'context compaction' feature will automatically summarise older context of conversations, letting Claude perform longer tasks without hitting limits.
Edited by Swetha Kannan