OpenAI released GPT-6 Astra with new prompting guidelines, advising developers to avoid 'slop words' and prioritize explicit instructions. It is the company's first major model update since GPT-5.6 Sol.
OpenAI reported GPT-6 Astra follows longer instructions better than its predecessors but is also more sensitive to context. That compares with GPT-5.6 Sol, which was less responsive to complex prompts.
GPT-6 Astra is built on OpenAI's latest architecture and targets collaborative tasks requiring sustained focus and initiative. Availability begins now, initially for developers and enterprise users.
"Delve," "foster," "leverage" and the rest of the slop words should be avoided," said OpenAI. Developers should infer the user’s intent and the scope of the task from the instructions and the conversation context so far.
The announcement follows OpenAI's push to improve model reliability and user control. OpenAI framed the update as a step toward more predictable and actionable AI behavior.
OpenAI did not say how GPT-6 Astra will perform on benchmark tests, and it raised concerns about unclear instructions in skill files. The company said developers should audit all skill files and context documents the model can access.
Source: thedecoder
