Anthropic 发布首个通用 Mythos 级模型 Claude Fable 5，内置对前沿 AI 开发的隐性安全干预
英文摘要
Anthropic released Claude Fable 5 (general availability) and Claude Mythos 5 (restricted), sharing the same underlying model with Fable 5 adding safety mitigations. The model achieves state-of-the-art on coding and agentic benchmarks, with a 1M-token context window and API pricing of $10/$50 per million input/output tokens. For sensitive topics like cybersecurity and biosecurity, queries are transparently routed to Opus 4.8; for requests targeting frontier LLM development, Anthropic silently reduces effectiveness via prompt modification, steering vectors, and PEFT without notifying users, affecting ~0.03% of traffic. This hidden intervention sparked widespread criticism from researchers and open-source advocates as anti-competitive and undermining trust. Fable 5 is temporarily included in subscriptions until June 22, after which it will require usage credits.
中文摘要
Anthropic 发布了 Claude Fable 5（全面可用）和 Claude Mythos 5（受限访问），两者基于同一底层模型，但 Fable 5 增加了安全保护措施。该模型在编码和智能体基准测试中达到顶尖水平，支持 100 万 token 上下文，API 价格为输入/输出每百万 tokens 10/50 美元。对于网络安全和生物安全等敏感话题，请求会被透明地路由到 Opus 4.8；而对于针对前沿 LLM 开发的请求，Anthropic 会通过提示修改、引导向量和参数高效微调等方式无声地降低模型有效性，且不通知用户，估计影响约 0.03% 的流量。这种隐性干预引发了研究者和开源倡导者的广泛批评，认为其反竞争且破坏信任。Fable 5 在 6 月 22 日前临时包含在订阅中，之后将需使用积分。
关键要点
Claude Fable 5 is the first generally available Mythos-class model, sharing the same base as the restricted Mythos 5 with additional safety guardrails.
Claude Fable 5 是首个全面可用的 Mythos 级模型，与受限的 Mythos 5 共享相同基座，但增加了额外的安全防护措施。
Pricing is $10/million input tokens and $50/million output tokens, with 1M context window, and it leads multiple coding/agentic benchmarks (e.g., 72.9% on CursorBench, 88% on Terminal-Bench 2.1, 80.3% on SWE-Bench Pro).
API 价格为每百万输入 tokens 10 美元、输出 tokens 50 美元，支持 100 万上下文窗口，并在多项编码/智能体基准测试中领先（如 CursorBench 72.9%，Terminal-Bench 2.1 88%，SWE-Bench Pro 80.3%）。
For cybersecurity, bio/chemistry, and distillation requests, Fable 5 transparently falls back to Opus 4.8; for frontier LLM development requests, it silently degrades capability via hidden interventions affecting ~0.03% of traffic.
对于网络安全、生物/化学及蒸馏相关的请求，Fable 5 透明地降级到 Opus 4.8；而对于前沿 LLM 开发请求，它会通过隐性干预无声降低能力，影响约 0.03% 的流量。
The hidden RSI suppression (prompt modification, steering vectors, PEFT) caused significant backlash from the AI community as anti-competitive and a threat to open research.
隐性的 RSI 抑制（提示修改、引导向量、参数高效微调）引发了 AI 社区的强烈反对，被认为是反竞争且威胁开放研究的行为。
Fable 5 is included in Pro/Max/Team/Enterprise plans until June 22, then will require usage credits due to capacity, reflecting the high inference cost of frontier models.
Fable 5 在 6 月 22 日前包含于 Pro/Max/Team/Enterprise 计划，之后因容量限制将需使用积分，反映了前沿模型的高推理成本。