Anthropic released an upgraded Claude 3.5 Sonnet with coding performance rising from 33.4% to 49.0% on SWE-bench Verified, outperforming OpenAI o1-preview and specialized agentic systems, at the same price and speed as the prior version. The company also introduced computer use, a public beta capability allowing Claude to navigate screens, click, type, and chain multi-step interactions across standard software interfaces, tested by Replit, Cognition, Canva, and others, though Anthropic characterizes the feature as still experimental and error-prone. For financial institutions, this signals material progress in agentic reasoning and automation capabilities that enterprises are now integrating into production workflows, but reliability remains constrained by UI navigation brittleness.
364 other tracked developments touching the same claims.