GPT-6 Astra is the first model in a while that feels like more than a benchmark flex. The computer-use jump is the real story. Not smarter chat. Actual work on a machine: browsing, forms, software, multi-step tasks. That’s a different product category. Grok 4.7 is expected around mid-September. Different bet: larger pretrain plus a lot of SpaceX engineering data. If that holds, this becomes a real comparison on agents that do the work, not just describe it. The next two weeks will be loud. The only test that matters is messy real workflows, not launch-day screenshots.