Update: GPT-6 Astra hits Azure Foundry; long-context costs 2x
OpenAI's just-shipped frontier model is now callable in Azure too — Entra controls, a pricier long-context tier, full agent tool belt. Plus a Copilot shake-up.
Copy markdown
Now live in Microsoft Foundry
Astra has started rolling out through Foundry's Limited Access Program, giving Azure teams a second way to call it — with Microsoft Entra sign-in, role-based access, and no reserved capacity to buy. Access widens to more customers over the coming days.
Long context now carries a 2x tax
On OpenAI's API, short-context calls stay $10/$50 per million in/out, but the new long-context tier jumps to $20/$75. Azure's US Data Zone deployment adds roughly 10% on top of every rate — budget for it before you point big-window jobs at Astra.
The full agent tool belt ships in the box
Astra supports hosted shell, apply-patch, computer use, web and file search, code interpreter, image gen, Skills, and MCP, with reasoning effort dialable from low to max. Enough to wire a self-contained coding or ops agent without bolting on your own tools.
Rate limits, cutoff, and the ChatGPT rollout
Tier-1 API keys start at 500 requests and 500K tokens per minute; the knowledge cutoff is April 30, 2026. In ChatGPT, Astra reaches Plus, Pro, Business, and Enterprise plans over the coming days.
Elsewhere: GitHub is merging Copilot Chat into one history
From Sept 28, Copilot Chat on github.com, mobile, and the cloud agent fold into a single experience — and chat retention stretches from 28 days to your account's lifetime. The code-review default also moves from Lite to Balanced unless you opt out first.