GPT-6 Astra tops DDD benchmark for multi-step retrosynthesis, nearing specialist models

CatAstro_Piyush · x · 2026-09-12

Vlad Aladin reports that in their DDD Benchmark, GPT-6 Astra outperformed every other frontier model tested on multi-step retrosynthesis, coming remarkably close to specialist-level synthesis planning models. The task predicts synthetic routes for small molecules — the same planning their Phase 2 molecule Rentosertib, which showed age-reversing effects across 6 aging clocks, required.

Original post →

More from Models

Models channel →