OpenAI’s Cursor cutoff and four builder moves
Your AI stack is a supply chain now. One vendor move can force a product decision, while new routing, deployment, and model options give you more ways to keep shipping.
In today’s edition
- OpenAI proposes ending Cursor’s direct access to its models on November 12 after Cursor’s acquisition by SpaceX.
- Vercel says you can deploy an eve agent from a prompt, model choices, and MCP connections in one minute.
- GLM-5.3 is now open-weight on OpenRouter with a 1M context window and configurable reasoning effort.
- Replit’s Auto model routing claims the same output quality at 65% lower cost than its previous version.
- Google is rolling out Gemini Omni 1.1 Flash for more controllable, production-focused generative video work.
OpenAI proposes a Cursor access cutoff
Cursor users may need a model-plan B. OpenAI says it is ending its partnership with Cursor following Cursor’s acquisition by SpaceX, and its proposal would end Cursor’s direct access to OpenAI models on November 12.
- The proposed cutoff applies to Cursor’s direct access to OpenAI models.
- OpenAI explicitly called out developers who rely on its models inside Cursor.
- The timing gives teams a defined date to review coding workflows, dependencies, and fallback options.
What it means Audit any Cursor workflow tied to OpenAI models now, test substitutes before November 12, and avoid letting a single provider dictate your developer experience.
Source: OpenAI
Vercel pitches one-minute agent deployment
Got an agent idea and no appetite for deployment chores? Vercel says its eve flow lets you add a prompt, pick models and MCP connections, then deploy an agent in one minute.
- Start from v0 with a prompt rather than a blank project.
- Choose both models and MCP connections before deployment.
- The deployed agent is ready to chat.
- Your agent is backed by a Git repo you own.
What it means Use this as a fast path for testing a narrow agent idea, then keep the Git-backed project as your handoff point when the prototype earns real engineering time.
Source: vercel
GLM-5.3 lands as open weights
Long-context agents have another option on the menu. GLM-5.3 is now open-weight and available through OpenRouter, positioned for complex software engineering, long-horizon agents, and cybersecurity work.
- The model has a 1M context window.
- It offers configurable reasoning effort.
- OpenRouter lists complex software engineering, long-horizon agents, and cybersecurity as target workloads.
- Open weights give builders another deployment and model-selection path.
What it means Benchmark GLM-5.3 on your longest agent and coding tasks, especially where a 1M context window or configurable reasoning could simplify your architecture.
Source: OpenRouter
Replit adds automatic model routing
Model picking is becoming a product setting instead of a weekly debate. Replit now lets builders select Auto, which chooses the best model for each task automatically.
- Replit says Auto selects the best model for each task.
- Its testing reports the same quality output at 65% lower cost than the previous version.
- The feature is designed to move model selection out of the builder’s way.
What it means Try automatic routing on a contained workflow and compare quality, latency, and spend against your fixed-model setup before you build more manual routing logic.
Source: Replit
Google rolls out controllable video generation
Video generation is getting a more production-minded tune-up. Google DeepMind is rolling out Gemini Omni 1.1 Flash with an emphasis on control, faster iteration, and more polished output.
- The rollout is aimed at making generative video highly controllable.
- Google says the model is faster to iterate on.
- It is positioned for more polished, production-grade use.
- You can try it in Flow by Google and elsewhere.
What it means Test Gemini Omni 1.1 Flash for repeatable creative production tasks where tighter control and faster iteration matter more than generating a one-off clip.
Source: GoogleDeepMind