Anthropic Releases Claude Fable 5.2 and Mythos 5.2—Real-World Testing Shows High Generation Costs

On Monday, Anthropic launched two new models: Claude Fable 5.2 and the restricted-release Mythos 5.2. According to the company, this model series can directly translate brief instructions into executable software, physical object designs, and independently tackle complex research problems previously requiring specialized teams. In terms of output speed, the new models deliver up to a 30% improvement over their predecessors. In Anthropic’s internal FrontierCode 2 benchmark, Fable 5.2 scored 49.1%, significantly outperforming Fable 5.1’s 32.4%. Though both models share the same underlying architecture, Fable 5.2 implements stricter safeguards for high-risk requests—such as those involving hazardous biology or cybersecurity—while Mythos 5.2 is available exclusively to vetted institutions in relevant domains. API pricing is standardized at $8 per million input tokens and $40 per million output tokens.