GPT-6 Astra Is Both Incredible and Frustrating

OpenAI’s Astra model shows incredible promise in areas like game development, high-level strategic thinking, and computer use, enabling users to create 3D models and playable games with simple prompts. However, the model often hesitates to complete tasks, preferring to explain rather than execute, and struggles to reliably follow specific instructions within AI skills. Consequently, the author plans to use Astra for creative and strategic tasks while sticking to a different model for execution-heavy work that requires strict adherence to established workflows.

• Astra excels at building 3D models and games using tools like Blender and Godot, allowing users to create games without learning complex software.
• The model demonstrates strong capabilities in high-level thinking and strategy, providing blunt and objective feedback on user plans.
• Astra can perform computer and browser-based tasks, such as drawing images using web applications.
• A significant drawback is Astra’s hesitation to directly complete tasks, often explaining the process instead of executing the command.
• The model also shows inconsistency in following specific instructions within AI skills, a problem not encountered with previous models like GPT-5.6 Sol or Claude Fable.
• The author will use Astra for game development, strategic thinking, and one-off computer tasks, but will continue using Sol Medium for execution-based work requiring adherence to AI skills.
  https://bender.layer3.press/articles/0f855061-9682-462f-8833-bc99e9de2806