Anthropic admitted they built an AI so capable they were scared to release it and the number that explains why is 250.
Milk Road AI Twitter · Milk Road AI (@MilkRoadAI) · 2026-05-16
Anthropic CFO Krishna Rao reveals in a clip that the company's unreleased model Mythos scored 250 on a coding benchmark where a prior frontier model had already been tested, prompting internal fear about releasing it.
Extraction
Topics: anthropicfrontier-modelsmodel-capabilitiesai-safety
Claims
- Anthropic built a model named Mythos that internal leadership considered too capable to release.
- Anthropic CFO Krishna Rao described Mythos achieving a score of 250 on a benchmark run against an open-source codebase.
- A previous frontier model had already been evaluated on the same open-source codebase before Mythos was tested.
- The number 250 is the key metric explaining Anthropic's hesitation to release Mythos.
Key quotes
Anthropic admitted they built an AI so capable they were scared to release it and the number that explains why is 250.
Anthropic's CFO Krishna Rao described in this clip what happened when they ran Mythos against an open source codebase that a previous frontier model had already...