Anthropic reveals Claude Opus 4.6, an enterprise-focused model with 1 million token context window for extended code capabilities

Anthropic has announced Claude Opus 4.6, a major upgrade to its flagship AI model focused on knowledge work and agentic coding tasks.

The model is intended to complete long, complex enterprise tasks in a single shot with higher-quality outputs across text, presentations, and spreadsheets than previous models released by Anthropic.

Code generation is a core focus for Claude Opus 4.6, which boasts a one million token context window – in beta – capable of using entire code bases for reference when generating code outputs.

Anthropic’s models are well-regarded for their raw capabilities at code generation, with Claude Opus 4.5 having set a new bar across coding benchmarks upon release in November 2025.

Claude Opus 4.6 achieves industry-leading scores across a number of popular benchmarks, including 65.4% in the coding benchmark Terminal-Bench 2.0 and 1,606 at GDPval-AA Elo, which tests performance for economically-valuable knowledge work.

Overall, the new model is also several points ahead of all other frontier AI models across enterprise-specific benchmarks including sector-specific tests for financial and legal capabilities.

Alongside Claude Opus 4.6, developers will also be able to use a new ‘Agent Teams’ capability as a research preview in Claude Code.

This allows multiple agents powered by Claude Opus 4.6 to work together in parallel, coordinating on a shared goal and reducing agentic processing bottlenecks.

API controls and better knowledge retrieval

Leveraging its large context window, Claude Opus 4.6 is intended to be far more useful for accessing information held in large datasets as well as long-form coding tasks such as bug detection and more complicated code implementation.

While competitor models such as Gemini 3 – as well as Anthropic’s own Claude Sonnet family of models – have boasted similar context windows in the past, this is a first for the Claude Opus family.

Anthropic said it has improved Claude’s ability to retain performance even when presented with a huge number of input tokens.

Tested with the ‘needle-in-a-haystack’ benchmark MRCR v2, which tests model fact-finding, attention, and reasoning across long and complex prompts, Claude Opus 4.6 scored 76% compared with Claude Sonnet 4.5’s 18.5%.

Alongside the new model, Anthropic announced additional controls for developers on the Claude API.

New controls include ‘adaptive thinking’, a configurable value that sets the degree to which Claude Opus 4.6 will use context clues to decide when its extended thinking capabilities are needed, as well as four ‘effort’ levels: low, medium, high (the default) and max.

Using this control, developers can reduce latency and inference costs when using the model for less complex problems or to power agents tasked with relatively simple workloads.

Long conversations in the Claude API will also now make use of context compaction, an offering in beta, which summarizes older conversational tokens to free up room in the context window over long back-and-forth tasks.

Context compaction happens when conversations hit a specific length, which developers can control.

Enterprise breakthroughs

Claude Opus 4.6 is intended as major leap for enterprise AI, with the largest improvements over its predecessor found in knowledge retrieval and sector-specific tool use.

In part, Anthropic is achieving this through new integrations for Claude directly within Microsoft 365, with improved data processing for Claude for Excel and new native slideshow support via Claude in PowerPoint.

In a demo video, Anthropic showed how Claude Opus 4.6 could be used to ingest enterprise spreadsheets and produce detailed competitor analysis, outputting new spreadsheets and an entire PowerPoint deck containing the most pertinent information.

Customer feedback on the new model emphasizes its strength across a wide range of complex enterprise tasks including codebase migration, agentic tool use, and extensive knowledge work.

“Claude Opus 4.6 handled a multi-million-line codebase migration like a senior engineer,” said Gregor Stewart, chief AI officer at SentinelOne.

“It planned upfront, adapted its strategy as it learned, and finished in half the time.”

Customers also praised the model’s extended context, particularly for knowledge work such as handling legal documents.

“Anthropic’s latest model represents a meaningful leap in long-context performance,” said Joel Hron, CTO at Thomson Reuters.

“In our testing, we saw it handle much larger bodies of information with a level of consistency that strengthens how we design and deploy complex research workflows. Progress in this area gives us more powerful building blocks to deliver truly expert-grade systems professionals can trust.”

Earlier this week, the legal giant suffered an 18% slump in its stocks in a selloff many investors linked to the launch of Anthropic’s enterprise-specific tool Cowork which can be used to automate some legal work.

“Claude Opus 4.6 excels in high-reasoning tasks, like multi-source analysis, across legal, financial, and technical content,” said Yashodha Bhavnani, head of AI at Box.

“Box’s eval showed a 10% lift in performance, reaching 68% vs. a 58% baseline, and near-perfect scores in technical domains.”

In practical tests, Claude Opus 4.6 also showed strong capabilities in managing IT issues as an autonomous agent.

“Claude Opus 4.6 autonomously closed 13 issues and assigned 12 issues to the right team members in a single day, managing a ~50-person organization across 6 repositories,” said Yusuke Kaji, general manager, AI at Rakuten.

“It handled both product and organizational decisions while synthesizing context across multiple domains, and knew when to escalate to a human.”

Claude Opus 4.6 is available at the same price as Claude Opus 4.5 via Anthropic’s API: $5 per million input tokens and $25 per million output tokens.