Astra-max claims vastly better intelligence-per-token even at low reasoning

adonis_singh · x · 2026-09-05

Blogger adonissingh continues his Astra-max evaluation thread: measured as intelligence per token, the model reportedly outperforms everything else even at low reasoning settings. Follow-up to his earlier eyebench-v3 post; unofficial third-party data.

Related event: Astra-max scores 95% on eyebench-v3 at half the cost of Sol-max(3 posts)→

Original post →

More from Models

Models channel →