Gemini 4 Carbon (high) tops MathArena at 85.4% expected performance, ahead of Fable 5.2 (max) at 82.8% and GPT-6.1 Sol (max) at 80.6%; their reported uncertainty ranges overlap.

MathArena’s model table captured November 13, 2026, showing expected performance, uncertainty and cost for Gemini 4 Carbon, Fable 5.2, GPT-6.1 Sol and GPT-6 Astra.

Published 13 Nov 2026 · 19:23 UTC

Jasper Dekoninck / MathArena ↗ · Source dated 13 Nov 2026

—

Browse all Live updates · Read Features