GPT-6.2 Astra scores 29.01% on Vals RSI-Index, still below Opus 5's ~33%

teortaxesTex · x · 2026-11-18

Vals AI's pre-release testing of GPT-6.2 Astra shows it is the highest-scoring OpenAI model they've tested, hitting 29.01% on the Vals RSI-Index (which measures autonomous AI R&D ability, with 50% calibrated as human-level) — an incremental gain over GPT-6 Astra's 27.07%. teortaxesTex notes the key takeaway: it's still "below Opus 5" (33%), with Fable 5.2 leading at 38%. The chart also positions Kimi K3, GLM 5.3, Gemini 3.7 Flash and Grok-4.7.

Original post →

More from Models

Models channel →