GPT-6.2 posts modest PowerPoint gains on IC Deck Bench, still trails Claude

peterjliu · x · 2026-11-18

Commenting on an IC Deck Bench chart (61 investment-committee deck tasks, run via Codex CLI at ultra effort and graded in PowerPoint), the author notes GPT-6.2's progress: GPT-6.2 Sol ultra scored 71.9, up 3.7 from GPT-6.1 Sol's 68.2, while GPT-6.2 Astra ultra hit 69.4, up 1.6 from 67.7.

Both remain well behind leader Compound auto (89.2), Claude Opus 5.5 max (86.1) and Claude Fable 5.2 high (83.6), sitting mid-pack overall.

Original post →

More from Models

Models channel →