Anthropic公布Fable 5.2与Mythos 5.2预发布评测
Anthropic公布了Fable 5.2和Mythos 5.2在电子硬件、化学、运筹学和理论物理四项预发布评测中的结果。公司称，Fable 5.2设计出三台可工作的硬件原型，并推进了两项理论物理计算；Mythos 5.2则为四个未发表的分子目标提出合成路线。在另一项运筹学测试中，Anthropic称Claude智能体刷新了MIPLIB中72道题的历史最佳结果，并为其中45道题证明了最优性。三台硬件原型尚未通过商业安规认证，四条分子路线也未经过湿实验检验。
这些评测显示Anthropic报告的模型能力已延伸至硬件、化学、运筹学和理论物理等任务。硬件成果仍需通过商业安规认证，分子合成路线也仍待实验室检验。
- Anthropic releases Fable 5.2 and Mythos 5.2 designed to help solve complex problems and ambitious projects
- Both offer a 20% cheaper token cost than previous versions
- The models are highly capable across coding, biology scientific and mathematics research
Anthropic’s latest Fable and Mythos releases have been unveiled, offering users superior outputs that are more accurate at a cheaper token cost than previous versions.
Fable 5.2 is available for all users and couples its improved inference with an improved set of cybersecurity and biology safeguards. But to get your hands on the expanded capabilities of Mythos 5.2, you’ll have to apply through Anthropic’s Life Sciences and Cyber Verification Programs.
Anthropic totes the new Fable and Mythos versions as being perfect for “ambitious, open-ended technical work” rather than as a “daily driver” — for which Anthropic recommends sticking to Opus 5.5.
Cheaper, quicker, and more accurate
The latest Fable release promises more bang for your buck over the 5.1 release. Anthropic says that 5.2 usage will sit at $8 per million input tokens and $40 per million output tokens — around 20% cheaper than the previous version.