A probe puts the frontier model near ten trillion parameters

Claim

A new knowledge-probe paper pegged GPT-5.5 at roughly 9.7 trillion parameters, with factual capacity still scaling log-linearly with compute even as reasoning saturates, while the same model topped a short-story creative writing benchmark.

Reported metric: ~9.7T parameters

Reporting date: 2026-04-30 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to April 30, 2026 (theinnermostloop.substack.com)