Matthew Berman reviews GPT-6 Astra using OpenAI benchmark claims and his own practical testsEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria.. He examines browser and computer useComputer use lets an AI agent interact with software through interface controls such as screens, buttons, keyboards and pointers., coding, game generation, three-dimensional scenes and knowledge work, including projects that require the model to plan and iterate over long contextsA context window is the maximum amount of tokenized information an AI model can consider during one processing session..
Astra produces strong one-prompt games and can sustain a multi-day simulation project, suggesting a meaningful gain in persistence and tool use. Matthew also notes that the model often falls back to familiar visual patterns and can leave an obvious AI-generated look even when the underlying functionality is strong.
The model costs $10 per million input tokens and $50 per million output tokens, so Matthew frames it as a premium system whose value depends on the task. He concludes that Astra expands what one agent can complete, while cost, design repetition and rollout constraints remain important limitations. Promotional requests are omitted.
Watch on YouTube




