Claude Opus 4.6 is a proprietary large language model released by Anthropic on February 5, 2026, succeeding Claude Opus 4.5 as the flagship of the Claude family. The release kept Opus-tier API pricing unchanged at $5 per million input tokens and $25 per million output tokens while extending the model's context window to 1 million tokens in beta, a first for the Opus line, with premium rates applying to prompts above 200,000 tokens.1
Capabilities
Opus 4.6 introduced "adaptive thinking," in which the model decides from context how much extended reasoning to apply, alongside explicit effort controls (low, medium, high, and max) that let developers trade test-time compute against latency and cost.1 The release also added context compaction, which summarizes older conversation history to sustain longer-running agentic tasks, and "agent teams" in Claude Code, where multiple agents work in parallel on parts of a problem, each with its own context window.1 On the applications side, Anthropic paired the launch with an updated Claude in Excel integration and a research preview of Claude in PowerPoint.1
Performance
Anthropic reported that Opus 4.6 achieved the highest score on the agentic coding benchmark Terminal-Bench 2.0 and led other frontier models on Humanity's Last Exam at release. On the company's GDPval-AA evaluation of economically valuable knowledge work, the model was developer-reported to outperform OpenAI's GPT-5.2 by roughly 144 Elo points and its predecessor Claude Opus 4.5 by about 190 points.1 As with earlier Opus releases such as Claude Opus 4.1 and the models of the Claude 4 series, these figures come from Anthropic's own announcement materials and had limited independent verification as of February 2026.
Release and access
The model launched simultaneously on claude.ai, the Claude API (as claude-opus-4-6), and major cloud platforms.2 An accompanying system card described safety evaluations and stated that the model's safety profile was comparable to that of Opus 4.5.3 Opus 4.6 arrived amid rapid iteration across the industry, competing with contemporaries such as GPT-5.1 and its successors, Gemini 3, and open-weight challengers like Kimi K3; within Anthropic's own lineup it sat above Claude Sonnet 4.5 and Claude Haiku 4.5 until later Claude 5-generation releases such as Claude Sonnet 5.