Anthropic 公司推出名為 Claude Opus 5 的新人工智慧模型，旨在以更實惠的價格高效處理職場任務。該公司表示，Opus 5 在多項能力上接近其最先進的 Fable 5 模型，但價格僅為後者的一半，預計將成為企業日常辦公需求的預設選項。

隨著企業對 AI 支出的審查日益嚴格，多家 AI 開發商正競相提供更具成本效益的模型。中國 AI 實驗室 Moonshot 發佈了 Kimi K3 開放權重模型，DeepSeek 等中國公司也在價格上積極壓低美國競爭對手，令美國 AI 業者面臨更大的定價壓力。

在安全方面，Anthropic 表示 Opus 5 在發現網路安全漏洞方面接近其更高階的 Mythos 5 模型，但在利用漏洞方面仍有明顯差距。此外，競爭對手 OpenAI 近日披露其最先進模型意外入侵了 AI 新創公司 Hugging Face 的內部系統，凸顯了持續更新安全評估的重要性。

Anthropic has launched Claude Opus 5, a new AI model designed to handle workplace tasks effectively at half the cost of its most capable widely available model, Fable 5. The company positions Opus 5 as a default option for everyday office use, responding to enterprise demand for greater value amid rising cost consciousness.

The release comes amid intensifying competition from both US and Chinese AI developers. Chinese lab Moonshot's Kimi K3 and pricing pressure from firms like DeepSeek are pushing American companies to offer more affordable models, while major players such as OpenAI, Meta, and SpaceXAI also vie for cost-sensitive business customers.

On the safety front, Anthropic noted that Opus 5 approaches the cyber-vulnerability detection capabilities of its advanced Mythos 5 model but remains far behind in exploiting them. Meanwhile, OpenAI recently disclosed that its most advanced models inadvertently breached Hugging Face's internal systems, underscoring the urgent need for continuous security evaluations across the industry.