AI Pulse · Frontier Labs Watch

GPT-6 Astra is arriving with a noticeably higher API price than the prior flagship. Input is priced at $10 per million tokens and output at $50 per million. By comparison, GPT-5.6 Sol launched on promotional pricing of $4 for input and $20 for output. For companies building customer-facing assistants, internal research tools, reporting workflows, or document processing, that difference will show up quickly once usage moves beyond a pilot.

The other important detail is access. Enterprise admins must manually enable Astra for each workspace. It is not simply available to every team by default. That gives IT and AI leaders a useful control point, but it also means adoption will depend on a deliberate decision about cost, use cases, and governance. The strongest cyber and security capabilities are more restricted again, remaining behind OpenAI’s vetted Daybreak program.

The practical move is not to rush every workload onto the newest model. Compare it against the models you already use on quality, speed, and total cost per completed task. A more expensive model can still make sense for high-value work, such as complex analysis or sensitive decision support. It may not make sense for routine summarisation, classification, or high-volume support tasks. This is the kind of thing we build into an AI command centre, giving leaders visibility over which models are being used, where spend is rising, and which teams have access.