OpenAI halts new sign-ups for $200 monthly Pro plan

OpenAI, the developer of ChatGPT, will temporarily suspend new sign-ups for its top-tier Pro plan after a surge in demand for its new AI model, Astra. The move is aimed at easing the load on its systems.

Thibault Sottiaux, OpenAI's head of product and platform, said on X on Thursday (local time) that the company would temporarily halt new sign-ups for the $200-a-month Pro plan. This was done, he said, "so that current users can keep enjoying the best experience and continue using Astra," according to Yonhap.

"This plan is putting the greatest strain on the system," Sottiaux said. "We decided to take this minimal step so we can keep offering the widest range of services possible."

However, he added, "there is no impact on existing accounts."

OpenAI runs three paid ChatGPT tiers: the low-cost Go plan ($8 a month), the standard Plus plan ($20 a month), and the premium Pro plan.

The Pro plan itself is split into two tiers, priced at $100 and $200 a month.

The sign-up freeze applies only to the $200 tier, which carries the highest usage cap.

Sottiaux said OpenAI was "working to expand capacity as quickly as possible." He added that other plans besides the $200 Pro tier, as well as the pay-as-you-go application programming interface, remain available.

OpenAI has not disclosed specific figures — such as changes in new sign-ups, user numbers or daily token usage — since unveiling Astra. It also has not said how long the sign-up freeze will last.

OpenAI's Astra has cleared every stage of the "CAPTCHA" test used to verify that an internet user is human.

According to Yonhap, OpenAI developer Sharif Shameem said on X on Tuesday (local time) that Astra solved all 48 stages of the CAPTCHA test known as "I'm not a robot." He said this earned Astra what he called a "proof of humanity" certificate.

The "I'm not a robot" test includes tasks such as identifying distorted letters and telling apart Dubai Chewy Cookies from dogs in similar-looking photos. It also involves locating specific objects, such as traffic lights, within images.

Astra's ability to clear every stage of the test suggests it has advanced to a near-human level across multiple domains. These include spatiotemporal reasoning such as image recognition, contextual reasoning such as understanding instructions, and computer-use skills.

On the day of its release, Astra scored 61 points under the existing evaluation system (version 4.1), entering the rankings tied for fifth place. It trailed Anthropic's Claude Fable 5.1, which topped the list with 66 points. After the new evaluation system (version 4.3) was applied, Astra rose to a tie for first place with Fable 5.1.