First impressions of GPT-6 Astra from a limited number of users indicate a significant shift in OpenAI’s approach to artificial intelligence. This model is not a regular chatbot, but acts as a full-fledged computer operator, independently performing complex tasks through a network of specialized agents.
Astra can interact with browsers, spreadsheets, websites and desktop applications in much the same way as humans. The master agent breaks a complex problem into its parts and launches subagents to examine the solution in parallel, find errors, and adjust strategies. Thanks to this, models can not only write code, but also independently test the code, fix discovered issues and complete multi-step workflows.
This approach may place additional load on the native processor. Companies using Astra and proprietary data may need to run models in isolated virtual environments and containers, which require computing resources to set up and manage. In addition, the local computer will run test scripts, compile code, and maintain the infrastructure for agents to access company data.
Therefore, the proliferation of such systems will increase the demand for CPUs, especially in enterprise infrastructure, which may benefit Intel and AMD. We’ve seen how this demand has weakened the RAM market. At the same time, Astra partially hides its internal reasoning chain, so fully sublimating it into an open model may be much more difficult. If this approach does become the standard, it will become more difficult for China’s open weighting program to replicate OpenAI’s capabilities.