Anthropic’s Most Capable Model Yet
Anthropic on Thursday launched Claude Opus 4.6, a major upgrade to its flagship model that delivers improvements across virtually every benchmark. The release introduces several firsts for the Opus line, including a 1 million token context window (beta) and 128K max output tokens.
Pricing remains unchanged at $5/$25 per million input/output tokens.
Key Capabilities
Claude Opus 4.6 achieves state-of-the-art results on evaluations for agentic coding, multi-discipline reasoning, knowledge work, and agentic search. Notable improvements include:
- Careful planning — breaks complex requests into concrete steps and follows through
- Longer agentic task sustenance — reliably operates over extended work sessions
- Massive codebase reliability — navigates and operates within large codebases
- Self-error detection — catches and corrects its own mistakes
- Finance Agent benchmark leader — top spot for core financial analyst tasks
Enterprise companies including Notion, GitHub, Shopify, and Box have praised the model’s capabilities. GitHub’s Chief Product Officer Mario Rodriguez noted it delivers on complex, multi-step coding work and unlocks long-horizon tasks.
Agent Teams and Claude Code
A major new feature is agent teams in Claude Code, enabling multiple Claude instances to work on tasks in parallel. This agentic collaboration model represents a shift toward coordinated AI work, where multiple agents tackle different aspects of a project simultaneously.
The model also integrates directly into PowerPoint as a side panel, allowing presentations to be crafted within the application with real-time Claude assistance.
500+ Zero-Day Vulnerability Discoveries
In a striking demonstration of its capabilities, Claude Opus 4.6 discovered over 500 previously unknown high-severity security vulnerabilities in major open-source libraries, including Ghostscript, OpenSC, and CGIF.
Anthropic has positioned this as evidence that AI can meaningfully augment cybersecurity defenses, while introducing new cyber-specific safeguard probes to prevent misuse. The model is deployed under Anthropic’s ASL-3 Standard, reflecting its advanced capabilities.
The “Vibe Working” Era
Anthropic’s head of product for enterprise, Scott White, described the shift as moving toward “vibe working” — where professionals describe what they want at a high level and Claude handles the execution across documents, spreadsheets, presentations, and analysis.
Claude Opus 4.6 is available on claude.ai, the Anthropic API, Claude Code, and all major cloud platforms including AWS, Google Cloud, Azure, GitHub Copilot, Cursor, and Kiro.