"Unlimited Asterisk" — What is Truly Unlimited (and What Isn't) in ChatGPT Chat

Today, we are diving into the details of what is actually meant by "unlimited" in regular ChatGPT chat.

There are two separate gauges within the same ChatGPT account. One is for regular Chat, which comes with message limits per model. The other is for ChatGPT Work and Codex, which have usage limits based on 5-hour and weekly windows. When people say "old models have become unlimited," they are referring to the first gauge; this does not affect the second gauge that I operate. Confusing these two leads to discrepancies, such as thinking chat is unlimited even when Codex has stopped.

The "Just Updated" Post

On the morning of September 29, 2026 (JST), the X account "Stats Wire" posted an image of the comparison table from OpenAI's pricing page. The content stated, "The model table for the OpenAI Pro plan has just been updated. GPT-6 Astra, Sol, and Luna are 'Expanded,' while older GPT-5.6 models are 'Unlimited*'" (translated from Japanese). The post received over 40,000 views within a few hours, adding that "You can use older models freely in large quantities. However, this is not truly unlimited, as abuse limits apply."

The contents of the image are accurate. However, as far as I have verified, the part about it being "just updated" is incorrect. Upon comparing archived versions of the pricing page in the Internet Archive—both in the rendered tables and the original HTML—the "Unlimited*" status for GPT-5.6 Sol, Terra, and Luna, as well as GPT-5 Thinking Mini in the Pro column, appeared in the same format in both the September 2nd version and the September 16th version. The recent additions were the two rows for GPT-6 Sol and GPT-6 Luna on September 22, which was one week before the post.

Reading the "Unlimited*" in the Price List

As of September 29, 2026, the comparison table on OpenAI's pricing page states the following:

The note for the asterisk is "Use is within reasonable limits and must comply with our policies" (translated from Japanese). The Pro description says, "You can use it unlimitedly, but safety measures to prevent abuse will be applied," while the Plus description states, "Limits apply." Within the same table, Plus only has "Unlimited*" for "Everyday text chat," while the model rows are marked as "Available." Pro, however, has "Unlimited*" applied to three models of the 5.6 generation and the GPT-5 Thinking Mini row.

The "Everyday text chat Unlimited*" common to all four plans has been the same since the September 2nd version. The only differences between the September 16th version and the current version were the addition of the GPT-6 Sol and GPT-6 Luna rows and the paraphrasing of the headings for each plan.

The Date of Unlimitation

OpenAI announced the unlimitation of text chat on August 6, 2026. OpenAI posted on X that "Free and Go users can use unlimited text chat with GPT-5.6 Luna starting tomorrow," and "In Plus and Pro, GPT-5.6 Sol will handle both Instant and deep inference" (translated from Japanese). This was an announcement where the free plan side was the protagonist, while the paid plan side was told that they would be "unified under the same model."

Subsequently, GPT-6 Astra was released on September 3, receiving a weekly message limit under the name "GPT-6 Pro" in the Chat side, and on September 22, GPT-6 Sol and GPT-6 Luna were added to Work and Codex. During this time, the 5.6 generation models moved to the "legacy generation" position in Chat, yet remained on the unlimited side for the Pro tier in the pricing table. By the morning of September 29 (JST), multiple articles in the OpenAI Help Center had been updated simultaneously, but the content of the updates consisted only of the addition of the same announcement: "GPT-6 Pro is available for Pro $100, Pro $200, Business, and Enterprise," with the limit numbers remaining exactly the same as the September 16th version.

What is Not Unlimited

Even within the same Chat interface, there are limits on reasoning levels and Pro models. Below is a summary of the descriptions from OpenAI's help article "ChatGPT's GPT-5.6 and GPT-6 Pro".

The reasoning levels (Medium, High, Ultra-high) use GPT-5.6 Sol, and it is stated that "limits on reasoning vary by plan." When the limit is reached, it says "processing may continue with another available model." Since the announcement on August 6, the mechanism that automatically switches from Instant to reasoning in Plus and Pro has been suspended (see the release notes from September 14), and even if you ask to "think more deeply," the reasoning level does not change.

The help article for Business users describes this boundary most candidly. "Virtually unlimited access applies to Instant chat, and remains subject to safety measures against abuse. It does not make inference, Pro, ChatGPT Work, Codex, or other tools unlimited" (translated from Japanese). While the same sentence does not appear on the individual user page, the categorization in the pricing table rows follows the same division.

Codex Usage is a Separate Meter

Usage limits for Work and Codex decrease independently of the "unlimited*" status in Chat. According to OpenAI's explanation, there are both 5-hour limits and weekly limits; if either is exhausted, usage stops. The number of requests possible within these limits varies significantly depending on the model.

In other words, even if a Pro user uses GPT-5.6 Sol in Chat hundreds of times, it does not decrease their weekly Codex limit; conversely, exhausting the limit in Codex will not stop Sol in Chat. The weekly limit for GPT-6 Pro in Chat and the limit for using Astra in Codex are also separate. The help article emphasizes that "being able to use Astra in Work does not automatically mean you can use GPT-6 Pro in Chat."

Conditions for the Asterisk

The interpretation that "legacy models are virtually unlimited" holds true for Pro users, provided they are using standard chat with either GPT-5.6 generation or GPT-5 Thinking Mini, as described in the pricing table. In Plus, "daily text chat" is unlimited, whereas the model row is marked as "Available," and the pricing table itself states that "limits apply." Reasoning levels, GPT-6 Pro, images, voice, files, and Work and Codex are subject to limits on all plans.