OpenAI has removed the GPT-5.5 model from ChatGPT, ChatGPT Work, and Codex platforms less than 6 months after its official launch. The company is directing subscribers toward the newer GPT-6 family, led by the GPT-6.1 Sol model designed for agentic coding.
Rapid retirement less than six months after launch
Capabilities and pricing of GPT-6.1 Sol and the ultra-fast tier
Halted referral tracking parameters and publisher analytics confusion
Recorded traffic decline and official silence from OpenAI
Frequently asked questions
Rapid retirement less than six months after launch
On Wednesday, October 14, 2026, OpenAI permanently removed the “GPT-5.5” model from ChatGPT, ChatGPT Work, and the Codex programming platform, just under 6 months after the model was rolled out to users. Through this step, the company aims to steer paid subscribers toward using the newer “GPT-6” model family, spearheaded by “GPT-6.1 Sol” designed specifically for agentic coding tasks.
This change encompasses all subscription plans aimed at individual consumers, business plans, major enterprises, and the educational sector. Conversely, the discontinuation decision did not include OpenAI’s API, meaning developers who call the “GPT-5.5” model directly within their software applications can continue using it for the time being.
“GPT-5.5” was officially released on April 23, 2026; however, subsequent developments outpaced it within roughly 11 weeks when the “GPT-5.6” model family—which included the “Sol”, “Terra”, and “Luna” editions—arrived last July. Codex platform documentation initially advised users to replace “GPT-5.5” with “GPT-5.6 Sol” before the sunset deadline, but the company has since launched two newer generations of the “Sol” series.
Capabilities and pricing of GPT-6.1 Sol and the ultra-fast tier
OpenAI introduced “GPT-6.1 Sol” on September 29 as a direct upgrade to the “GPT-6 Sol” model. The company affirms that this enhanced release “nearly matches the intelligence level of the GPT-6 Astra model in agentic coding, computer use, and professional work, but at one-fifth the standard input and output token pricing of Astra.”
Regarding API pricing, using “GPT-6.1 Sol” costs $2 per million input tokens, just $0.10 per million cached input tokens, and $10 per million output tokens, making it an attractive economic option for software enterprises.
Paid ChatGPT subscribers can select the new model within the Codex and ChatGPT Work platforms, while it is not yet available in the standard chat interface. At the beginning of October, the company also began rolling out an “ultra-fast” tier that it claims generates code and text at roughly 8 times the usual rate. The company explained in a system card appendix that in internal Codex traffic simulations, “GPT-6.1 Sol” demonstrated a 56 percent reduction in high-severity incompatibility signals compared to the “GPT-5.6 Sol” model.
Halted referral tracking parameters and publisher analytics confusion
In a related context concerning the sixth-generation updates, the specialized publication Search Engine Roundtable reported in a separate piece that the ChatGPT platform appears to have stopped appending UTM marketing campaign tracking parameters to outbound links it generates. This sudden change complicates the task for publishers and website owners in determining the volume of traffic coming directly to them from ChatGPT conversations.
The report noted that this change appeared to coincide with the launch of the GPT-6 series about a week prior, keeping in mind that OpenAI had started adding tracking parameters to its outbound links last June to help sites measure traffic.
Cedric Hösle from Adobe wrote via LinkedIn explaining the technical aspect: “GPT-6 opens links using a new window command with noopener and no-referrer attributes, dropping the ChatGPT source tag. There is no tag and no referral reference, which is why your analytics software records the visit under direct traffic.”
Recorded traffic decline and official silence from OpenAI
To back up this observation with figures, Cedric Hösle examined data from roughly 500 web pages where ChatGPT accounted for at least 5 percent of total traffic, comparing statistics from the period spanning October 9 to 13 with the immediately preceding week.
The comparison results showed that visits officially attributed to ChatGPT dropped by about 31,000 visits per day across those pages, while at the same time, visits classified under the “direct traffic” category for the same pages rose by about 26,000 visits per day, confirming that traffic shifted without its true source appearing in dashboards.
Search Engine Roundtable noted that it remains unclear whether this change in the link-opening mechanism was intentional on the company’s part or resulted from an unintended software bug, while OpenAI has not issued any public comment clarifying its stance on the matter.
Frequently asked questions
Question: Which platforms did OpenAI remove the GPT-5.5 model from?
Answer: The model was removed from ChatGPT, ChatGPT Work, and the Codex platform across all individual, business, enterprise, and education plans, while remaining available to developers via the API.
Question: What distinguishes the new GPT-6.1 Sol model?
Answer: It matches the intelligence of GPT-6 Astra in agentic coding and computer use at one-fifth the price, and recorded a 56 percent reduction in serious incompatibility signals compared to GPT-5.6 Sol.
Question: What are the pricing rates for using GPT-6.1 Sol via the API?
Answer: The cost is $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens.
Question: How did the GPT-6 update affect website traffic statistics?
Answer: The system stopped sending tracking parameters and referral sources in outbound links, leading to a drop of about 31,000 daily visits attributed to ChatGPT in a sample of 500 pages, converting them into direct visits.
