Anthropic Unveils Claude Opus 4.6 with a 1M Token Context Window
Anthropic has launched Claude Opus 4.6, the first model in its Opus series to feature a massive 1 million token context window. This latest release brings significant improvements in programming, reasoning capabilities, and new productivity-enhancing features.
Enhanced Performance and Capabilities
According to the company, Opus 4.6 significantly surpasses its predecessor in several key areas:
- Advanced Programming: The model supports longer-running agentic tasks, handles large codebases more reliably, and is more adept at self-debugging and code verification.
- Benchmark Dominance: Opus 4.6 outperformed other top models in tests like BrowseComp (finding information online) and Terminal-Bench 2.0 (agentic coding). It also surpassed OpenAI's GPT-5.2 and Claude Opus 4.5 in the GDPval-AA benchmark for complex problem-solving.
Innovative Features and Integrations
Anthropic is also rolling out a suite of powerful new tools, many available in research or beta preview:
- Agent Teams: In Claude Code, users can now assemble teams of AI agents to collaborate autonomously on complex tasks.
- Deeper Integrations: Claude now handles complex tasks in Excel more effectively and introduces a new integration with PowerPoint (available for Max, Team, and Enterprise subscribers).
- Adaptive Thinking: This feature allows Claude to automatically decide when to engage in deep reasoning for more complex queries. Developers can also manually set the model's effort level.
- Context Compaction (Beta): To manage the large context window, this function automatically summarizes older context, enabling Claude to handle extended tasks without hitting memory limits.
With the ability to assemble 'teams of agents' and new integrations for Excel and PowerPoint, Claude is becoming an even more powerful tool for complex, real-world tasks.
Availability and Pricing
Claude Opus 4.6 is now available on claude.ai, through the API, and across all major cloud platforms. The API pricing remains unchanged at $5 per million input tokens and $25 per million output tokens.