GPT-6 Astra Pro/Ultra review: first LLM to nail one-shot prompts nearly every time
rschu · x · 2026-09-09
After using GPT-6 Astra Pro/Ultra on several complex tasks, rschu says it's the first LLM that correctly understands and executes one-shot prompts almost every time, crediting part of that to the Desktop harness and testing capabilities. He concludes it deserves the hype — but pushes back on AGI claims: "AGI, c'mon."
More from Models
- Apodex 1.1 launches with open-weight 35B mini model for local deployment — mhdfaran · 2026-09-09
- Apodex open-sources FrontierAgent agent framework and ships 35B open-weight Apodex 1.1 mini — mhdfaran · 2026-09-09
- Accusation: a covert industry packages and sells your work identity to AI labs — kevinafischer · 2026-09-09
- Community Releases Uncensored 27B Fine-tune Using Marchenko-Pastur Noise Distillation — EvilEnginer · 2026-09-09
- ChatGPT Astra Won a Pokémon Battle Autonomously — at $25 Per Match — bucketbot91 · 2026-09-09
- Reddit users slam Claude's opaque usage percentages, demand metered billing — NTXL · 2026-09-09