Finally official 5.6 Pro & Astra Web Limits

[miuuyy]

Limits

OpenAI has finally published the official limits for GPT-6 Pro and GPT-5.6 Sol Pro.

The most important part is simple: Chat, ChatGPT Work, and Codex will keep separate allowances.

Many people worried that OpenAI would merge them into one shared limit. I never thought that was likely, and thankfully, it did not happen.

This is also a good outcome for Codex Web GPT. The repository is approaching 4,300 stars, so OpenAI certainly knows it exists. At the same time, the project has not become associated with abuse or quota bypassing. That is exactly where I want it to stay.

The new Zero Risk mode makes this even better. It does not read or modify ChatGPT and does not automate prompt submission. You send the prepared prompt manually while keeping the Codex context and harness. This removes the browser-automation risk entirely for users who prefer that workflow.

Thank you to everyone who starred the repository, tested releases, and reported bugs.

The limits are reasonable

On the $200 Pro plan:

• GPT-6 Pro has 200 messages per week.
• GPT-5.6 Sol Pro has a separate allowance of 170 messages per day.
• Combined usage across both models is limited to 200 messages per day.

Honestly, 170 GPT-5.6 Sol Pro messages per day is a lot.

This is still useful even if you do not use the model constantly. You can bring GPT-6 Astra or GPT-5.6 Sol Pro into Codex for an architecture discussion, code review, difficult bug, consultation, or several focused edits with the full project context and Codex harness.

What about Plus?

OpenAI did not include Plus in the table, so this part is only my prediction.

I expect Plus to remain close to its current behavior: GPT-6 Astra may be available on the web only through Work, while GPT-5.6 remains available in regular Chat.

We will know for sure when the rollout is complete.

My real usage

For the last seven days, my research pipeline has continuously used Pro models without any noticeable change.

It made around 700 Pro-model sends in one week. Only after that did cooldowns become frequent enough to make the workflow temporarily unusable.

My conclusion is that the practical limits have not changed dramatically. OpenAI probably still uses additional internal cooldowns or rolling limits that are not shown in the table.

What comes next

When GPT-6 Astra becomes available, Codex Web GPT will add explicit model-version selection. You will be able to choose the exact model, such as GPT-6 Astra or GPT-5.6 Sol Pro, instead of using one generic Pro route.

I also plan to add usage metrics to the launcher. Successful turns will be recorded locally, giving you a clearer estimate of your recent usage and when a cooldown may be approaching.

Overall, this is a good result: the limits are now documented, Chat and Work allowances remain separate, and Zero Risk mode provides a browser-automation-free way to use ChatGPT models with the Codex harness.