Anthropic 于 11 月 9 日正式推出两款全新 AI 模型,分别命名为面向普通用户的 Claude Fable 5.2 和仅向通过审核的专业机构开放的 Claude Mythos 5.2 1。在价格与性能方面,Fable 5.2 的 API 调用费用较前代版本下调了 20%,具体输入价格为每百万 Token 8 美元,输出价格为每百万 Token 40 美元,而生成速度相比 5.1 版本最高提升 30%,若开启“快速模式”则可提升至 2.5 倍 1。目前,Fable 5.2 已在 AWS、Google Cloud 和 Microsoft Azure 等主流云平台上线,而 Mythos 5.2 则需要用户通过 Cyber 或 Life Sciences 验证计划申请方可使用 1。
在能力展示上,预发布测试揭示了新模型的显著突破,其中 Fable 5.2 协调超过 1,000 个子智能体,在 8 天内从零开始成功构建了一个原型操作系统 1。此外,该模型还在约 1,000 个代理协作下,于 30 小时内完成了费马大定理的 Lean 形式化证明,并解决了黑洞物理及全息对偶领域的开放科学问题 1。相比之下,专业版 Mythos 5.2 则在对 34 个腺相关病毒衣壳进行建模的任务中,成功完成了其中 28 个的构建 1。为了应对潜在风险,新模型增强了安全对齐机制,不仅拦截渗透测试等高风险任务,还加强了反蒸馏保护,使数据泄露减少了 90% 至 99% 1。
Anthropic officially announced the release of two new AI models, Claude Fable 5.2 and Claude Mythos 5.2, on November 9 1. The announcement distinguishes between user access levels: Claude Fable 5.2 is designed for general users, while Claude Mythos 5.2 is restricted to professional institutions that pass a verification process 1. Currently, Claude Fable 5.2 is available on major cloud platforms including AWS, Google Cloud, and Microsoft Azure, whereas access to Claude Mythos 5.2 requires an application through specific Cyber or Life Sciences validation programs 1.
The update includes significant cost reductions and performance enhancements for Claude Fable 5.2 1. Compared to the previous version 5.1, API call fees have decreased by 20%, with input costs now set at $8 per million tokens and output costs at $40 per million tokens, down from $10 and $50 respectively 1. Generation speed has improved by up to 30% compared to version 5.1, and can increase to 2.5 times faster when "fast mode" is enabled 1. Pre-release testing demonstrated the model's capabilities in complex tasks, such as coordinating over 1,000 sub-agents to build a prototype operating system from scratch in eight days 1. Additionally, approximately 1,000 agents completed a Lean formalization proof of Fermat's Last Theorem within 30 hours 1.
Beyond coding tasks, the new models show advanced abilities in scientific research and security 1. Claude Fable 5.2 successfully addressed open problems in black hole physics and holographic duality 1. In the field of life sciences, Claude Mythos 5.2 modeled 28 out of 34 adeno-associated virus capsids 1. To enhance safety, Anthropic has strengthened alignment measures and anti-distillation protections, which reportedly reduce data leakage by 90% to 99% 1. The systems also intercept high-risk tasks, such as penetration testing 1.
评论
还没有评论,欢迎留下第一条。