Anthropic 推出新版人工智慧模型 Fable 5.1，在程式编写与科学研究任务上的能力显著提升，能更有效处理大型软体专案、程式审查、实验设计与复杂图表读取。为了回应客户对使用成本的诉求，新模型大幅调降处理既有快取资讯的费用达 75%，并优化 token 使用效率，从而显著降低企业用户的整体运行成本。

除了性能升级与降价，Anthropic 也改进了安全分类器以减少误报，提升资安与生物领域用户的工作效率。同时，该公司调整了数据保留政策，允许企业客户在使用其顶级 AI 模型时将数据保留在自有的云端架构中而非 Anthropic 端，并预计同步向限定用户推出更新版的邀请制模型 Mythos 5.1。

面对来自中国开源低成本模型的竞争压力，Anthropic 与 OpenAI 正积极拓展涵盖金融、医疗、资安等专业领域的 AI 工具，以争取更多企业客户并支撑其庞大估值。据报导，Anthropic 的年化营收预计将突破 650 亿美元，且已提交保密上市申请，最快可望于今年秋季公开挂牌交易。

Anthropic is releasing a new version of its AI model, Fable 5.1, which delivers enhanced capabilities in coding and scientific tasks, including complex software projects, code reviews, experiment design, and interpreting dense diagrams. To address customer concerns regarding AI operating expenses, the company is reducing the pricing for reprocessing previously handled information by 75% while improving token efficiency to substantially lower overall usage costs.

Alongside performance upgrades and cost reductions, Anthropic has enhanced its safety classifiers to reduce false positives, allowing users in cybersecurity and biology to work more smoothly. The company is also updating its data policy to allow enterprise clients to retain data on their own cloud infrastructure when using advanced models, and it plans to release an updated version of its invite-only model, Mythos 5.1, to select users.

Facing competition from rapidly improving, low-cost open-weight models from China, Anthropic and rival OpenAI are intensifying efforts to build specialized tools across professional industries like healthcare and finance to court business clients. Anthropic is on track to surpass $65 billion in annualized revenue and has filed confidential paperwork to go public, with a market debut anticipated as early as this fall. (Key numbers: 650)