Claude Fable 5 发布，基于 Mythos 模型并新增安全防护
英文摘要
Anthropic has released Claude Fable 5, which shares the same underlying model as Mythos but incorporates additional safeguards. Andrej Karpathy reports that benchmarks show state-of-the-art performance by a wide margin, and qualitatively the model represents a major version bump comparable to the step change from Claude 4.5. It excels in long, difficult problem-solving sessions, enabling users to tackle far more ambitious tasks such as generating explainers, dashboards, and custom single-use apps. The safeguards are currently overactive and may need tuning, and the model retains some quirks. Karpathy sees this release as a transformative shift that will dramatically increase demand for on-demand software creation.
中文摘要
Anthropic 发布了 Claude Fable 5，该模型与 Mythos 使用相同的基础模型，但增加了额外的安全防护措施。Andrej Karpathy 指出，基准测试显示该模型以显著优势达到最先进水平，定性来看这是一个重大版本跃升，堪比以前 Claude 4.5 的进步。它在长时间解决高难度问题方面表现突出，允许用户执行更具雄心的任务，如生成解释器、仪表板和一次性自定义应用。安全防护目前过于敏感，可能需要后续调优，模型仍存在一些怪癖。Karpathy 认为此发布将推动软件开发的变革，极大刺激按需软件创作的需求。
关键要点
Claude Fable 5 shares the same underlying model as Mythos but adds new safeguards.
Claude Fable 5 与 Mythos 使用相同基础模型，但新增了安全防护。
Benchmarks show state-of-the-art results with a large margin over competitors.
基准测试以明显优势达到最先进水平。
Qualitatively, it is a major version bump on par with the Claude 4.5 leap, especially for long, complex problem-solving.
定性评估显示这是一个重大版本进步，堪与 Claude 4.5 的飞跃相媲美，尤其在长时间复杂问题解决中。
The safeguards are initially over-triggered and will likely be tuned, while the model still exhibits some quirks.
安全防护初始过度触发，后续可能调优，模型仍有怪癖。
Enables generation of elaborate software artifacts (explainers, visualizers, dashboards, single-use apps), signaling a shift in software demand.
支持生成复杂的软件产物（解释器、可视化、仪表板、一次性应用），预示着软件需求的范式转变。