Anthropic 发布 Claude Fable 5.2 与 Mythos 5.2
Anthropic Launches Claude Fable 5.2 and Mythos 5.2
Anthropic 发布 Claude Fable 5.2 和 Claude Mythos 5.2，在编码、推理和视觉方面有所改进，输出生成速度最高提升 30%。Fable 5.2 在中等努力设置下处理高难度任务时使用的 token 少于 Fable 5.1，且相比 5.1 拦截次数减少约 55%。两款模型共用同一底层模型，公开版 Fable 5.2 带有网络安全和生物领域防护，Mythos 则通过 Anthropic 的生命科学与网络验证计划及企业版 Claude Security 提供。
推荐理由
AI 评分 70
原文给出了两款新模型在编码、推理和视觉上的改进，以及相对 Opus 5.5 的定位与选型建议。
主题
正文
Anthropic today introduced its latest models, Claude Fable 5.2 and Claude Mythos 5.2, which have improvements in coding, reasoning, and vision.
Anthropic says Fable 5.2 and Mythos 5.2 communicate more clearly than the 5.1 versions, with outputs generated up to 30 percent faster. Fable 5.2 uses fewer tokens than Fable 5.1 on challenging tasks when set to medium effort.
The company suggests using Fable and Mythos for "the most ambitious, open-ended technical work." Anthropic recommends starting with Opus 5.5 for most coding and knowledge tasks, then switching to a model like Fable or Mythos that can handle more complex requests, or stepping down to a more affordable model like Sonnet or Haiku 5.5.
Benchmarks suggest Fable and Mythos 5.2 outperform Opus 5.5 for computer use, knowledge work, cybersecurity, agentic scientific research, agentic coding, multidisciplinary reasoning, and more.
Both use the same underlying model, but the public-facing Fable 5.2 model has cybersecurity and biology safeguards to prevent misuse. Compared to Fable 5.1, Fable 5.2 has approximately 55 percent fewer blocks. Mythos does not have the same limitations as Fable, and it is available through Anthropic's Life Sciences and Cyber Verification Programs as well as through Claude Security on Enterprise plans.