OpenAI details GPT-6 Astra quality fixes: misfiring legacy skills, experiment hit 4-5k users
rohanpaul_ai · x · 2026-09-13
Addressing reported quality issues with GPT-6 Astra, OpenAI's Thibault Sottiaux posted a reset and postmortem of fixes:
- Some skills written for previous models were triggering too often or preventing the model from checking its work;
- An opt-in context-management experiment caused early stops or replies to older messages — now disabled, with an estimated 4-5k users affected;
- Badly configured engines that measurably degraded quality for a long tail of traffic were removed.
The team says additional minor improvements have shipped and the experience should feel significantly better.
More from coding & agent
- GPT-6 Astra playtests games like a human via in-app browser, reading DOM to save tokens — cedric_chee · 2026-09-13
- Same LLM eval, different scores: a method to tell real regressions from noise — bgoncalves · 2026-09-13
- Spend 20% of your time having AI build your app, 80% using it — Daniel_Farinax · 2026-09-13
- OpenAI shows off GPT-6 Astra community builds, from 3D anatomy to Unity air traffic sims — OpenAIDevs · 2026-09-13
- Professor publishes annotated transcript on using AI agents to teach the perceptron — davidbau · 2026-09-13
- GitHub repo curates 500+ agentic AI projects for ideas and implementation study — Roger_M_Taylor · 2026-09-13