Anthropic has rolled out Claude Sonnet 4.6, positioning it as a high-performance mid-tier model that matches or exceeds flagship Opus 4.6 in several key domains — at just 20% of the cost.

Benchmark performance: On SWE-Bench Verified, Sonnet 4.6 scored 79.6%, just shy of Opus 4.6’s 80.8%, despite running at a fraction of the price. Anthropic also reported that Sonnet 4.6 outperformed Opus 4.6 on certain agentic financial analysis and office-task benchmarks — marking a first for the mid-tier Claude line.

Developer preference: Early Claude Code testers preferred Sonnet 4.6 over its predecessor 70% of the time, and even chose it over Opus 4.5 in 59% of head-to-head comparisons.

Computer-use leap: Sonnet’s OSWorld score jumped dramatically from under 15% in late 2024 to 72.5%, highlighting rapid gains in autonomous computer-use capabilities.

Why it matters: Anthropic appears to be accelerating a trickle-down strategy, compressing the gap between premium and mid-tier performance in record time. As pricing pressure intensifies globally — especially from lower-cost Chinese competitors — Sonnet 4.6 strengthens Anthropic’s position in the high-volume, agentic AI deployment layer where cost-performance balance is critical.