Claude Opus 4.6

• 🏢 Anthropic flagship reasoning model, released February 2026.
• 📏 One million token context window, 128K max output.
• 🧠 Adaptive thinking dynamically decides reasoning depth per task.
• 🔧 Built for agentic coding, long-horizon and autonomous workflows.
• 👁️ Multimodal vision, function calling, and web search supported.
• 📚 Reported 90.2% on Anthropic's BigLaw Bench legal reasoning.
• 🔒 Improved robustness to prompt injection over Opus 4.5.
• ⚡ Context compaction summarizes old tokens for long sessions.

Anthropic

Anthropic is an American artificial intelligence company headquartered in San Francisco, focused on developing large language models under the Claude family. Founded with a strong emphasis on AI safety research, the company has become one of the most…

Claude Opus 4.6 is Anthropic's flagship reasoning model, released in February 2026, positioned for complex software engineering, agentic workflows, and long-horizon autonomous tasks. It is the direct successor to Claude Opus 4.5 and arrives alongside the smaller, more cost-effective Claude Sonnet 4.6 and a latency-tuned Claude Opus 4.6 Fast variant.

Compared with Opus 4.5, Anthropic describes 4.6 as bringing enhancements in reasoning quality, instruction-following, and overall performance, with the model planning more carefully at the start of tasks and revisiting its reasoning before settling. A key generational change is the expanded one-million-token context window with 128K max output, versus the 200K context used in Opus 4.5 evaluations. On Anthropic's reported MRCR v2 long-context retrieval benchmark, Opus 4.6 scores 76%, and the company reports 65.4% on Terminal-Bench 2.0 and 90.2% on BigLaw Bench. On the provider's SWE-bench Verified, it reaches 81.42% with a prompt modification.

The release also introduced adaptive thinking, letting Claude decide when and how much reasoning to apply, plus context compaction that summarizes older context as token limits approach during long agentic runs.

Opus 4.6 is not the newest in its family. It was followed by Claude Opus 4.7 and later Claude Opus 4.8, which Anthropic now lists as its most capable generally available model.