AI Search tests GPT-6 Astra across coding, desktop software and creative productionAgent evaluation measures how well an AI agent performs intended tasks across defined, repeatable conditions.. A custom ray-tracing scene and an Unreal Engine game illustrate how a separate critic can review results and steer revisionsA critic agent is an AI component that reviews another agent's work and provides feedback for correction or improvement., although neither example reaches its requested quality score within the initial round limit. The game still needs further work to correct misplaced scene elements.
Computer-use demonstrationsComputer use is an AI capability that interprets a graphical interface and operates software through actions such as clicking, typing, and selecting. include drawing in a browser editor, making a sprite animation, playing a virtual piano and composing music in a digital audio workstation. Other tasks reconstruct a property in Blender and produce an explained mathematics animation. Several outputs improve after additional prompts, supplied assets or revised narrationIterative refinement improves an AI result through repeated cycles of building, inspection, feedback, and revision., making the demonstrated workflow a collaboration with feedback rather than consistently finished work from one instruction.
The strongest demonstrations involve extended tool use, but the review also shows repeated failures to locate a hidden frog and weak results on another visual-classification exercise. Research and benchmark examples broaden the comparison without independently proving accuracy across all tasksTask accuracy is the degree to which an AI system produces correct results for a defined task.. The practical takeaway is to assess working outputs and iteration costs alongside headline benchmark claims.
Watch on YouTube
