OpenAI: training new internal model with 'unprecedented' math benchmark performance

TheZvi · x · 2026-09-09

OpenAI says it has been training a new internal model since August 28 that has shown "unprecedented performance" on its benchmarks including mathematics, with training ongoing and performance still improving. Zvi notes this ongoing capability ramp is the real story — bigger than the Millennium Prize solution itself.

Original post →

More from Models

Models channel →