Anthropic 发布 Claude Fable 5 与 Mythos 5：相同基座模型，不同安全防护，全新 Mythos 级别
英文摘要
Anthropic released two models, Claude Fable 5 and Claude Mythos 5, on June 9, 2026. Both belong to the new Mythos class, positioned above the Opus tier, and share the same underlying model. Fable 5 is generally available with safety classifiers that fall back to Opus 4.8 on flagged requests, while Mythos 5 has lifted cyber safeguards and is limited to Project Glasswing. The models offer a 1M-token context window and 128k output tokens, priced at $10/M input and $50/M output. Anthropic reports Fable 5 achieves state-of-the-art results across nearly all benchmarks, including software engineering, finance, vision, and long-context tasks, with Stripe demonstrating a 50-million-line code migration in one day. Classifiers activate in under 5% of sessions, and over 95% of Fable sessions experience no fallback, effectively matching Mythos 5 performance.
中文摘要
Anthropic 于 2026 年 6 月 9 日发布 Claude Fable 5 和 Claude Mythos 5 两款模型。二者同属 Mythos 级，位于 Opus 级之上，且共用相同基座模型。Fable 5 面向一般用户开放，内置安全分类器，在标记请求时回退至 Opus 4.8；Mythos 5 则移除了网络安全防护，仅通过 Project Glasswing 有限开放。模型支持 100 万 token 上下文窗口和 12.8 万 token 输出，定价为每百万输入 token 10 美元、每百万输出 token 50 美元。Anthropic 称 Fable 5 在几乎所有基准测试中达到最先进水平，涵盖软件工程、金融、视觉和长上下文任务；Stripe 利用该模型一天内完成了一个 5000 万行代码库的迁移。分类器在不到 5% 的会话中触发，超过 95% 的 Fable 会话无需回退，此时性能与 Mythos 5 相当。
关键要点
Fable 5 and Mythos 5 share the same underlying model; the only difference is safeguard classifiers.
Fable 5 和 Mythos 5 共用相同基座模型，唯一区别在于安全分类器。
Fable 5 achieves state-of-the-art on nearly all tested benchmarks, including software engineering (50M-line migration in one day), finance, vision, and long-context tasks.
Fable 5 在几乎所有测试基准上达到最先进水平，包括软件工程（一天完成 5000 万行代码迁移）、金融、视觉和长上下文任务。
Safety classifiers fall back to Opus 4.8 for flagged cyber, bio/chem, and distillation requests; triggers occur in under 5% of sessions.
安全分类器对标记的网络、生物化学和蒸馏请求回退至 Opus 4.8，触发率低于 5%。
Both models offer a 1M-token context window and 128k output tokens, priced at $10/M input and $50/M output, less than half the cost of Mythos Preview.
两款模型均提供 100 万 token 上下文窗口和 12.8 万输出 token，定价为每百万输入 10 美元、输出 50 美元，较 Mythos Preview 便宜一半以上。
Mythos 5 remains limited to Project Glasswing with strongest cyber capabilities; it ran novel autonomous genomics research for a week and produced preferred scientific hypotheses.
Mythos 5 仅通过 Project Glasswing 开放，具备最强网络能力，并自主进行了一周的新颖基因组学研究，产出了在盲评中受到青睐的科学假设。
Over 95% of Fable sessions experience no classifier fallback, effectively matching bare Mythos 5 performance.
超过 95% 的 Fable 会话无需分类器回退，实际性能与无安全防护的 Mythos 5 相当。