Anthropic, the operator of the artificial intelligence (AI) chatbot Claude, on the 17th released Claude Sonnet 4.6, which raises the performance of its midtier flagship model Sonnet following its top-end model Opus. It is a follow-up update 12 days after the unveiling of Opus 4.6 on the 5th this month.
Anthropic said Sonnet 4.6 is a model that improves overall capabilities across coding, computer use, long-context reasoning, agent planning, and knowledge-work tasks. In particular, it supports a 1 million token (1M) context window, targeting demand from corporations to process long-form materials such as large-scale codebases, contracts, and reports at once.
In the free and Pro plans, Sonnet 4.6 is applied as the default model, and API pricing remains at $3–$15 per 1 million tokens, the same as the previous version. The company said Sonnet 4.6 is immediately available on Claude, CoWork, Code, the API, and major cloud platforms.
It also narrowed the gap with Opus 4.6 in performance metrics. Sonnet 4.6 posted 79.6% on SWE-bench Verified and 72.5% on OSWorld-Verified, approaching Opus 4.6 (80.8%, 72.7%) in each. It scored 1,633 on GDPval-AA, surpassing Opus 4.6 (1,606), and recorded 63.3% on the Finance Agent benchmark, higher than Opus 4.6 (60.05%). Anthropic said initial tests showed developers preferred Sonnet 4.6 over Sonnet 4.5 by about 70%, and even compared with Opus 4.5, preference was 59%.
In the industry, attention is on the possibility that the downward diffusion of Opus-level performance will take off in earnest and reorganize an expense structure centered on high-priced models. As Anthropic has recently been expanding enterprise automation tools in succession, there is also an outlook that raising the performance of midtier models could further accelerate the pace of AI-based task replacement.