Overview
- Anthropic on Monday released Claude Fable 5.2 for broad use and Claude Mythos 5.2 for organizations approved through its Cyber and Life Sciences verification programs.
- The company says both models produce outputs up to 30% faster than the 5.1 generation and it cut standard API rates by about 20% to $8 per million input tokens and $40 per million output tokens.
- Anthropic published internal demonstrations showing advanced agentic work, including coordinating more than 1,000 subagents to build a bootable prototype operating system and formalizing a proof in a machine‑checkable form.
- To balance capability and risk, Anthropic reports expanded safety evaluations, new post‑launch monitoring and Enterprise Frontier Safeguards entering private beta with early customers and an option for zero data retention for eligible clients.
- The launch is timed as Anthropic readies an IPO and follows outside scrutiny over possible recursive self‑improvement and containment testing, which has prompted extra controls and closer regulatory attention.