Anthropic发布Fable 5模型，面向智能体长时间任务的Mythos公开版
英文摘要
Anthropic released Fable 5, the public version of its Mythos model previously restricted under Project Glasswing. It is available on Pro, Max, and Team plans, free until June 22 after which usage credits apply. The model is designed for multi-hour agentic sessions, autonomously spinning up sub-models, gathering data, and writing and testing its own code. Hard safety blocks on cybersecurity, biology, and chemistry cause it to fall back to Opus 4.8 when triggered. A Reddit user is asking for community feedback on its real-world performance and weak points.
中文摘要
Anthropic发布了Fable 5模型，这是此前仅在Project Glasswing中向特定合作伙伴提供的Mythos模型的公开版本。该模型可在Pro、Max和Team计划上使用，6月22日前免费，之后需消耗使用额度。它专为长时间智能体任务设计，能自主启动子模型、收集数据，并自写自测代码，支持数小时的连续运行。模型在网络安全、生物和化学等领域设有硬安全限制，触发后回退至Opus 4.8。一位Reddit用户正征集社区的实际使用反馈，探讨优缺点。
关键要点
Anthropic publicly released Fable 5, the former restricted-access Mythos model.
Anthropic公开了Fable 5，即此前受限的Mythos模型。
Free access on Pro/Max/Team plans until June 22; afterwards usage credits are required.
在6月22日前，Pro/Max/Team计划用户可免费使用，之后需消耗使用额度。
Optimized for long agentic tasks, it can autonomously spawn sub-models, gather data, and write/test code over hours.
专为长时间智能体任务优化，能自主创建子模型、收集数据并编写/测试代码，持续数小时。
Hard safety restrictions on cybersecurity, biology, and chemistry topics cause fallback to Opus 4.8.
在网络安全、生物和化学话题上设有硬性安全限制，会回退至Opus 4.8。
A Reddit thread invites user experiences to gauge real-world effectiveness and overhyped aspects.
Reddit帖子征集用户实际体验，以评估其真实能力与炒作之处。