Anthropic has officially released Claude Opus 4.6, its most advanced large language model designed to handle the most complex and agentic tasks. Announced on February 5, 2026, the new model introduces a massive 1-million-token context window and specialized “agent teams” that allow multiple AI units to coordinate on large-scale projects. The release marks a significant push into professional sectors like legal, finance, and software engineering, where high-precision reasoning is paramount.
A new frontier for agentic workflows
Claude Opus 4.6 is built specifically for “long-horizon” work tasks that require an AI to plan, execute, and verify results over extended periods. Unlike previous iterations that focused on single-prompt responses, Opus 4.6 excels at autonomous multi-step processes.
A core innovation in this release is the “agent teams” feature. This allows Claude to spin up multiple subagents that work in parallel, mimicking a real-world engineering team. In a notable internal experiment, Anthropic demonstrated this capability by having 16 Claude agents build a functional C compiler from scratch in just two weeks, a feat that would typically require months of human effort.
Unprecedented context and adaptive thinking
Anthropic has addressed the memory constraints of modern AI by introducing a 1-million-token context window in beta. This allows the model to ingest entire codebases or thousands of pages of legal documents in a single pass without losing coherence.
The model also debuts “adaptive thinking,” a feature that allows the AI to dynamically decide how much reasoning effort to apply to a specific query.
- 128K Output Limit: Opus 4.6 can generate up to 128,000 tokens in a single response, doubling the previous limit.
- Adaptive Effort: Users can control the depth of thinking via a new “effort” parameter, balancing performance against cost and latency.
- Enhanced Vision: The model features improved visual reasoning for analyzing complex charts, diagrams, and technical layouts.
Leading benchmarks in professional domains
In terms of raw performance, Claude Opus 4.6 has set several industry records. It reportedly outperforms OpenAI’s GPT-5.2 on benchmarks related to real-world knowledge work, particularly in the finance and legal sectors.
According to external evaluations, the model achieved a 90.2% score on the “BigLaw Bench,” a rigorous test for legal reasoning. In the “GDPval-AA” evaluation, which measures office task efficiency, it scored 1606 Elo, surpassing its predecessor by 190 points and signaling a major leap in its ability to handle “economically valuable” professional work.
Security and global availability
To support high-stakes enterprise workflows, Anthropic has introduced several administrative controls. The new “compaction API” provides server-side context summarization, effectively enabling infinite conversations by compressing older data while retaining essential context.
Data residency has also been prioritized. The new “inference_geo” parameter allows companies to specify where their model inference runs, with US-only routing available for industries with strict compliance requirements.
Claude Opus 4.6 is available immediately on the Claude Developer Platform, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Pricing for the model starts at $5 per million input tokens and $25 per million output tokens, maintaining the same base rate as its predecessor.
For more information on the recent release, please visit the official Claude Opus 4.6 announcement.
