GPT-6 Astra shows the best instruction-following since Opus 4.5
0xSammy · x · 2026-09-08
@realmcore calls GPT-6 the best instruction-following model since Opus 4.5, more steerable than 4.6, 4.7, 4.8, fable, 5.5 and 5.6. @menhguin adds two observations switching from fable to Astra: it abides by tiny contradictions in a 30k-token system prompt, and it willingly runs for many hours without explicit prompting.
More from Models
- Researcher says GPT-6 built a cell model that explained how Ozempic works in an hour — DeryaTR_ · 2026-09-08
- Nvidia CEO declares "AGI has arrived" after GPT-6 Astra, reigniting definition debate — mrtac96 · 2026-09-08
- Artificial Analysis Intelligence Index v4.3 raises private-question weight to 45% — ArtificialAnlys · 2026-09-08
- GLM-5.3 and Kimi K3 lead open weights at 44, still 9 points behind GPT-6 Astra and Claude Fable 5.1 — ArtificialAnlys · 2026-09-08
- Local vision models manage ~3Hz tracking; real-time agent-level speed still a year out — yacineMTB · 2026-09-08
- Taming Qwen3 27B's Over-Reasoning With an AI-Generated System Prompt Patch — ea_man · 2026-09-08