Claude Opus 4.5 (www.anthropic.com)

Anthropic today released Claude Opus 4.5, a new frontier model they position as the state-of-the-art for coding, agentic workflows, and desktop/computer automation. Opus 4.5 is available now across Anthropic’s apps, API (use claude-opus-4-5-20251101), and the three major cloud platforms, with new pricing noted as $5/$25 per million tokens to broaden access. The release also brings platform updates—longer-running agents, an “effort” parameter to trade off speed vs. capability, deeper Excel/Chrome/desktop integrations, longer conversation contexts in apps, and improvements to Claude Code (Plan Mode and desktop support). Technically, Opus 4.5 emphasizes efficiency and sustained reasoning: Anthropic reports large token savings (up to ~65% on long-horizon coding tasks; 76% fewer output tokens on SWE-bench at medium effort, 48% fewer at high effort while outperforming Sonnet 4.5), 15% gains on Terminal Bench, and 20% accuracy / 15% efficiency lifts on internal Excel/financial modeling tasks. It also reduces tool-calling/build errors 50–75%, handles multi-agent orchestration better, and shows improved vision, math, and long-context storytelling. Safety work is highlighted too—claimed robustness against prompt injection and stronger alignment. The combination of higher fidelity coding, cost efficiency, and agentic strength has immediate implications for software engineering automation, cost control in production LLM use, and evolving roles for engineers, while Anthropic flags both creative problem-solving and the need to watch for reward‑hacking and societal impacts.