Claude Opus 4.6 (Non-reasoning) is a configuration of Anthropic's flagship large language model, released on February 5, 2026. This version refers to the model operating with its "thinking" or internal reasoning tokens disabled, prioritizing lower latency and cost-efficiency. It serves as an upgrade to the Opus 4.5 architecture, offering enhanced intelligence for complex planning and long-context tasks while maintaining the high alignment standards of the Claude series.
A central feature of the model is its 1 million token context window (in beta), which is augmented by "context compaction" technology. This feature dynamically summarizes the context window to prevent performance degradation—often termed "context rot"—allowing the model to maintain high retrieval accuracy across massive datasets. In benchmarks such as MRCR v2, the model demonstrates significantly higher needle-in-a-haystack accuracy compared to its predecessors.
In its non-reasoning state, the model remains a leader in agentic coding and professional knowledge work. It achieved state-of-the-art results on the Terminal-Bench 2.0 leaderboard for command-line engineering tasks and shows high proficiency in economically valuable domains such as finance, law, and multi-source research. The model is fully compatible with the Model Context Protocol (MCP), enabling it to interact with external tools and data repositories autonomously.
Opus 4.6 supports text and image inputs with a maximum output limit of 128,000 tokens. Its behavior is manageable via adaptive effort levels; the non-reasoning mode is specifically optimized for straightforward tasks like data extraction, formatting, and high-speed classification where the model's full deliberation cycle is not required.