On February 5, 2026, Anthropic released Claude Opus 4.6 — the most powerful model in the Claude family. With major improvements in coding, reasoning, and a new agent teams feature, Opus 4.6 sets a new standard for what AI models can do in real-world workflows.
What's new in Opus 4.6
Agent teams
The headline feature is agent teams — the ability to split complex tasks into parallel subtasks, each handled by a specialized agent. This is a game changer for developers working in large codebases, where multiple files and systems need to be updated simultaneously.
In Claude Code, agent teams coordinate autonomously: one agent can refactor a module while another writes tests and a third updates documentation — all in a single session.
1M token context window
For the first time in an Opus-class model, Opus 4.6 supports a 1 million token context window (in beta). This means entire codebases, legal documents, or research papers can be processed in a single prompt.
The improvement is not just about size — retrieval quality is dramatically better. On the MRCR v2 benchmark (8-needle, 1M variant), Opus 4.6 scores 76%, compared to just 18.5% for Sonnet 4.5.
Adaptive thinking
Opus 4.6 introduces adaptive thinking — the model automatically determines when deeper reasoning is needed, without the developer having to configure it. For finer control, there are four effort levels: low, medium, high (default), and max.
This means faster responses for simple tasks and deeper analysis for complex ones, optimizing both speed and cost.
Benchmark results
Opus 4.6 doesn't just improve — it leads across nearly every major benchmark:
- Terminal-Bench 2.0: Highest score among all models for agentic coding tasks
- Humanity's Last Exam: #1 on complex multidisciplinary reasoning
- GDPval-AA: Outperforms GPT-5.2 by ~144 Elo points on economically valuable tasks
- BrowseComp: Best performance for locating hard-to-find information online
- BigLaw Bench: 90.2% accuracy with 40% perfect scores in legal reasoning
- Life sciences: Nearly 2x better than Opus 4.5 in computational biology, organic chemistry, and phylogenetics
What this means for developers
Better coding, longer sessions
Opus 4.6 plans more carefully, sustains agentic tasks for longer, and operates more reliably in larger codebases. Code review and debugging have seen significant upgrades.
Context compaction
During long-running tasks, the model can automatically summarize older context to stay within limits while retaining key information. This is especially useful for extended coding sessions.
128K output tokens
With support for up to 128,000 output tokens, Opus 4.6 can generate substantial code, documentation, or analysis in a single response.
Office integration
Anthropic expanded Claude's integration into productivity tools:
- Claude in Excel: Enhanced performance for complex, multi-step spreadsheet operations
- Claude in PowerPoint: A research preview that enables design-aware presentation creation directly within PowerPoint
Safety and alignment
Opus 4.6 maintains safety parity with Opus 4.5, which was already Anthropic's most-aligned frontier model. It shows the lowest over-refusal rate among recent Claude models, meaning it's both safe and practical to use.
Pricing and availability
Claude Opus 4.6 is available on claude.ai, the API, and all major cloud platforms. Pricing remains at $5 per million input tokens and $25 per million output tokens.
Conclusion
Claude Opus 4.6 represents a significant leap forward in AI capabilities. The combination of agent teams, a 1M context window, and adaptive thinking makes it a powerful tool for software development, research, and business operations.
If you're building AI-powered products or want to integrate Claude into your workflows, we'd love to help you get started.