Last week I wrote about the model race — GPT-6 Astra, Claude Fable, Cursor, Grok. This week the story got sharper. OpenAI is winding down direct model access for Cursor after SpaceX bought Anysphere, with a proposed shutoff on November 12, 2026. Astra will not ship through Cursor. And Anthropic did not wait: more Claude compute inside the editor, Fable 5.1 already live, and higher Claude Code limits rolling in.
If you build for clients the way I do — WordPress, Breakdance, Headspin UI, agency work where the IDE is half the product — this is not gossip. It is a supply-chain event. Model access is a lease, not a purchase.
On August 28, OpenAI notified SpaceX that it intends to cancel the contract that feeds OpenAI models into Cursor. The company is using a change-of-control window after SpaceX closed its roughly $60 billion acquisition of Cursor’s parent on August 14. OpenAI says it cannot be confident SpaceX will stay inside its terms of service, citing prior disputes with Musk-controlled companies. The proposed cutoff is November 12. During the wind-down, OpenAI will not provide future models — including GPT-6 Astra — to Cursor.
Cursor CEO Michael Truell said OpenAI models are about 5% of Cursor user traffic and that talks continue. That number sounds small until you remember how teams actually work: a minority of sessions can still be the ones where people reach for a specific model on purpose. If your muscle memory is “open Cursor, pick GPT,” you have a calendar problem, not a vibe problem.
Anthropic moved immediately. Tom Brown said Cursor has been a trusted partner since Sonnet 3.5 and that Anthropic will keep increasing compute for Claude inside Cursor. Claude Fable 5.1 is already available in the editor — Cursor’s own notes put it at the top of CursorBench 3.2 in their testing, with strong “verify your own work” behavior on long agentic tasks. Separately, Claude Code weekly limits get a permanent 25% bump starting September 14 (with the temporary 50% boost still on until then).
That is the practical story for me this week. Astra is the flashy AGI-era headline on the OpenAI side — computer use, coding, science, cybersecurity, Critical threshold, Daybreak gating. Fable 5.1 is the model I can actually point at a Breakdance tree or a client theme without waiting for Trusted Access. Same $10/$50 per million token pricing band as Astra on the API side, but day-one availability across Claude’s platforms and inside Cursor beats a waitlist for my shipping calendar.
Here is the boring, durable plan I’m using on my own site work and client projects:
The Musk–Altman story will get the clicks. The builder lesson is quieter. SpaceX now owns Cursor, Colossus-scale compute is in the background, and Composer exists so Cursor can treat frontier APIs as substitutable inputs. OpenAI is protecting Astra’s terms and cyber posture. Anthropic is buying developer mindshare with capacity and Fable. None of that is abstract if your billable hours depend on an autocomplete pane.
I am not abandoning OpenAI. I am refusing to let one contract clause strand a week of client work. Astra can be best-in-class on OSWorld, FrontierMath, and ExploitBench and still be the wrong default if it is not in the tool I open every morning. Fable can win the day by being present, cached cheaper on reads, and tuned for agentic coding.
Between now and November 12 I’ll watch three things: whether OpenAI and Cursor reverse or soften the cutoff; how hard Anthropic’s extra compute lands under real Cursor traffic; and whether Astra’s ChatGPT/API rollout makes a separate OpenAI lane feel seamless enough that losing it inside Cursor barely hurts. If you ship WordPress and UI for a living, run that same checklist. Pick a primary, name a backup, and do not confuse a great model with guaranteed distribution.
The race is not only who scores highest on a bench. It is who shows up inside the editor when the contract changes.