xAI Claims Grok Never Silently Downgrades Intelligence, Gets Mocked

himanshustwts · x · 2026-08-13

xAI recently released the model card for Grok 4.6, detailing its capabilities across various evals for coding, engineering, and knowledge work, alongside pre-deployment safety testing and their safeguard stack. The card explicitly claims that they "never silently downgrade intelligence or fall back to other models."

However, this assertion was met with mockery online. A user laughed off the claim, implying that the industry is well aware of the common practice among AI providers to secretly throttle model capabilities or route requests to cheaper models in the background.

Related event: xAI Faces Transparency Controversy Over Grok 4.6 Release(3 posts)→

Original post →

More from Fun

Fun channel →