Rumor claims Kimi faked performance by serving Claude; DeepSeek new model surprises in evals

realsohamparekh · x · 2026-09-11

An unverified claim alleges Kimi was faking its performance by serving Claude responses to users. Soham Parekh notes this still doesn't explain why self-hosted Kimi models perform well, and adds that DeepSeek's recent model rollout amusingly outperformed even Astra and Fable in his team's evals. The rumor's authenticity is uncertain, but it raises questions about model identity verification and benchmark trustworthiness.

Original post →

More from Fun

Fun channel →