Environment
- Plan: ChatGPT Plus ($20/mo)
- Feature/Model: Codex (GPT-6 Astra Light / Desktop Agent Control)
- Target App: Blender 3D (GUI & Viewport Control)
Summary of Issue
The quota drain for Codex using GPT-6 Astra during desktop application integration is critically flawed. Within 2 to 5 minutes of starting a basic session—before any actual modeling task was rendered or completed—100% of the 5-hour usage limit was instantly exhausted, resulting in an immediate mid-task lockout.
Steps to Reproduce
- Launch Codex in the ChatGPT Desktop app under a Plus plan.
- Initiate a task requiring desktop integration/screen control in Blender.
- Allow the agent to scan the viewport and perform initial environment checks.
- Observe the usage cap drop from 100% available to 0% within 2–5 minutes.
Expected Behavior
A standard 5-hour usage allowance on a paid tier should allow for reasonable active work sessions. Visual screen captures and UI automation should not burn through an entire multi-hour quota in minutes without clear warning, token budget toggles, or rate-limit protections.
Actual Behavior
The session abruptly ends with “You’ve hit your usage limit” mid-execution. The feature is functionally unusable in its current state for desktop/3D workflows under standard plans.
Please review the Astra visual token calculation and provide rate-limit transparency before launching automated UI features to paid tiers.