Anthropic officially unveiled Claude Sonnet 4.6 on February 17, 2026, marking what the company calls its most capable Sonnet model to date. According to Anthropic, Sonnet 4.6 delivers performance comparable to Claude Opus 4.5 — the company's frontier model released in November 2025 — while maintaining the more accessible Sonnet pricing of $3 per million input tokens and $15 per million output tokens. The model is now the default across Anthropic's consumer products, including claude.ai and Claude Cowork, and is also available through the API and major cloud platforms.
One of the most notable advancements in Sonnet 4.6 is its significantly improved computer use capability. When Anthropic first introduced a general-purpose computer-using model in October 2024, it openly acknowledged that the feature was "still experimental — at times cumbersome and error-prone." Sixteen months of iteration have produced a dramatically different picture. On OSWorld, the standard benchmark for AI computer use, Sonnet models have shown steady and consistent gains. Early Sonnet 4.6 users report human-level performance on tasks such as navigating complex spreadsheets and completing multi-step web forms across multiple browser tabs. Importantly, the model also demonstrates a major improvement in resistance to prompt injection attacks — a key security concern for agentic AI systems — outperforming its predecessor Sonnet 4.5 and performing on par with the higher-tier Opus 4.6.
In the coding domain, results from Claude Code testing underscore the significance of this release. Users preferred Sonnet 4.6 over Sonnet 4.5 approximately 70% of the time, citing improvements in context awareness before code modification and reduced duplication of logic. More strikingly, users preferred Sonnet 4.6 over Opus 4.5 in 59% of comparisons — praising it for being less prone to overengineering, more consistent in instruction following, and more reliable across multi-step tasks. Leading developer platforms including GitHub Copilot, Cursor, Replit, and Bolt have all confirmed strong real-world performance, with some highlighting specific breakthroughs in large codebase navigation and production-ready output quality.
From a technical standpoint, Sonnet 4.6 introduces a 1 million token context window in beta, enabling it to hold entire codebases, extensive contracts, or dozens of research papers within a single request. The model supports both adaptive and extended thinking modes and benefits from context compaction in beta, which automatically summarizes older context as conversations lengthen. A reliable knowledge cutoff of August 2025 gives it a temporal edge over other models in its family. On the safety front, Anthropic's evaluations concluded that Sonnet 4.6 is "as safe as, or safer than" its recent predecessors, with no significant concerns around high-stakes misalignment — a finding the company emphasizes as integral to every new model release.