OpenAI Paused ChatGPT Pro ($200): GPT-6 Astra Is Still on the API

OpenAI paused new $200 ChatGPT Pro subscriptions on 10 September 2026. GPT-6 Astra demand is the reason. The API is not paused.

Thibault (Tibo) Sottiaux, who leads ChatGPT and Codex products at OpenAI, wrote:

If you wanted the ChatGPT Pro UI, new seats are closed until OpenAI says otherwise. If you wanted GPT-6 Astra in Cursor, Codex, Claude Code, or your own app, that path is the API. Model ID gpt-6-astra is live on RouterPlex at OpenAI list rates, on the same prepaid key as Claude, Gemini, and the rest of the catalog.

What OpenAI actually paused #

This is admission control on one ChatGPT SKU, not an Astra shutdown.

The day before, Tibo warned that Astra demand was "really unprecedented" and that OpenAI might pause new Pro subscriptions if it continued. The 10 September post is that pause. Do not read it as cancelling people who already pay $200.

Astra launched 3 September 2026. One week later the highest individual ChatGPT plan stopped taking new money so existing users could keep the model. That is a capacity story, not a model withdrawal.

The API is still open #

Tibo named the API in the same sentence as the other plans. Metered tokens are the release valve: you pay for what you run, OpenAI can add capacity without selling another unlimited-feeling $200 seat.

That is also why a ChatGPT Pro pause is the wrong reason to stop shipping Astra in a product. ChatGPT Pro is a subscription chat product. The API is a token bill. Cursor, Codex CLI, and your backend never needed the $200 SKU. They needed gpt-6-astra and a key.

RouterPlex is the second half of that sentence: one prepaid key, vendor list price, 0% markup, $0 top-up fee, balance never negative, hard per-key budget. When the key hits the cap, the request stops. It does not overdraft. It is not ChatGPT.

ChatGPT Pro vs the Astra API #

Do not convert $200/month into "unlimited Astra." OpenAI did not say Pro was unlimited. It said Pro accounts put the most strain on the system. A fixed-price seat that can run long agent threads is expensive for the provider even when it is expensive for you.

A 40,000-in / 2,000-out Astra turn is about $0.50 at the list rate. The 272K cliff still applies: one fat context doubles the request. That is the failure mode a $200 seat hides and a budgeted key makes visible.

If you wanted Pro for the ChatGPT app, wait for OpenAI. If you wanted Astra for work, buy tokens.

Use GPT-6 Astra on RouterPlex #

Model ID gpt-6-astra. OpenAI SDK base URL https://api.routerplex.com/v1. Anthropic SDK and Claude Code: https://api.routerplex.com.

curl https://api.routerplex.com/v1/chat/completions \
  -H "Authorization: Bearer $ROUTERPLEX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6-astra",
    "messages": [
      {"role": "user", "content": "Review this diff like a $200 Pro seat is not required."}
    ]
  }'

import os
from openai import OpenAI
 
client = OpenAI(
    api_key=os.environ["ROUTERPLEX_API_KEY"],
    base_url="https://api.routerplex.com/v1",
)
 
response = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[{"role": "user", "content": "Turn this stack trace into a one-line fix plan."}],
)
print(response.choices[0].message.content)

Cursor: override the OpenAI base URL to https://api.routerplex.com/v1 and add gpt-6-astra. Codex CLI: custom provider with wire_api = "responses". Same catalog ID.

Put Astra on its own key with a small budget until the first usage row looks right. A loop that would have been "someone else's $200 seat" is your prepaid balance here.

Same key for any model, not only Astra #

The Pro pause is an OpenAI ChatGPT-SKU problem. It is not a reason to run only OpenAI.

On RouterPlex the same key reaches GPT-6 Astra, GPT-5.6 Sol, the OpenAI hub, the Anthropic hub, Gemini, Qwen, Kimi, Grok, and the rest of the catalog. Swap the model string. Keep the budget.

That is the practical answer to "Pro is paused." You do not need a $200 ChatGPT seat to call Astra. You do not need a second vendor account to fall back to Claude when Astra is the wrong tool.

Start a $5 RouterPlex test, put gpt-6-astra on a budgeted key, and send one live request. Existing ChatGPT Pro users keep their seats. Everyone else still has the API.