What is so attractive about Opus 4.6 that makes technology people so interested?

With adaptive thinking, conversation reduction, and a range of leading performance indicators, Claude Opus 4.6 is a great improvement over Opus 4.5. As Anthropic calls it, they upgraded their smartest model. Here it is Outstanding features of Claude Opus 4.6.

Claude Opus 4.6 brings Anthropic’s most advanced inference capabilities to Microsoft Foundry, where intelligence and reliability combine to automate work. In Foundry, Opus 4.6 can enable knowledge from anywhere: leverage Foundry IQ to access data from M365 Work IQ, Fabric IQ, and the web. This model is best applied to complex tasks in programming, knowledge work, and agent-driven workflows, supporting deeper inference while providing superior instruction following to ensure reliability.

Now, developers can delegate their most complex work to AI systems for full lifecycle development, from requirements gathering to deployment and maintenance. Business users can create documents, perform research, and author content with polish and in-depth industry understanding.

Claude Opus 4.6 is considered the best Opus model for programming, business development and professional work. With a context window of 1 million tokens (beta version) and a maximum output of 128k tokens, Opus 4.6 is ideal for:

• Production code
• Complex agents
• Office tasks
• Financial analysis
• Network security
• Use a computer

By combining Anthropic’s most advanced modeling with Foundry’s comprehensive toolset, customers can move from testing to production faster, without having to pair infrastructure.

A new level for automatic programming

Opus 4.6 handles large codebases well and is especially effective with long-term tasks such as refactoring, error detection, and complex deployments.

This allows senior engineers to assign work that previously took days, including everything from requirements gathering to deployment and maintenance, while remaining focused on review and decision-making.

With Foundry’s managed infrastructure and operational controls, teams can shorten development times from days to hours, while maintaining the rigor needed for real production systems.

Improve knowledge work efficiency across the enterprise

Opus 4.6 delivers a breakthrough in the way businesses approach knowledge work, defined by Anthropic based on three pillars: search, analysis and creation. In Microsoft Foundry, these capabilities can be applied directly to enterprise data, workflows, and agent-driven applications.

Opus 4.6 understands the conventions and norms of specialist fields, creating documents, spreadsheets and presentations that look and read like the work of an expert. Combined with Foundry’s access control and governance capabilities, this makes Opus 4.6 well-suited for finance, legal and other high-precision industries where quality, context and compliance are critical.

Improve your ability to use computers and delve into details

According to Anthropic, Opus 4.6 delivers major improvements in computing usage, with strong performance above industry standards for image understanding and multi-step navigation. Opus 4.6 sets a new standard for computing. Claude can now operate the computer more precisely, handle more complex tasks, and work seamlessly across multiple applications.

It can interact with software, navigate interfaces, complete forms, and move data between applications.

Agents, security, and reasoning in high-risk situations

Opus 4.6 is also suitable for agent-based workflows, reliably orchestrating complex tasks across dozens of engines. It can proactively instantiate child agents, parallelize work, and promote task development with minimal supervision.

For security workflows, Opus 4.6 provides deep inference capabilities, allowing teams to identify subtle patterns and complex attack vectors with high accuracy.

Adaptive thinking and effort

There are two features of Claude Opus 4.6 that determine whether it needs to use expansive thinking and how much effort it puts into that thinking process.

Adaptive thinking allows the model to determine the complexity of your request. Based on simplicity or complexity, it will decide whether to use expansive thinking or not. Instead of manually setting the number of tokens to use for this, Claude will adjust his budget based on the complexity of each request.

The effort parameter allows you to set Claude’s level of willingness or caution in spending tokens. Essentially, it means you can strike a balance between token efficiency and response completeness.

When using Claude Opus 4.6 in the API, you can set these parameters manually. For example:

• Maximum Effort: Claude always uses an expansive mindset and has no constraints on depth.
• High Effort: With this default setting, Claude is always thinking and providing deep reasoning.
• Medium Effort: This activates moderate thinking and it can skip thinking for the simplest queries.
• Low Effort: Claude forgoes thinking for simple tasks and minimizes thinking in favor of speed.