OpenAI CEO Sam Altman on Wednesday unveiled a collection of updates to ChatGPT, giving customers extra management over the corporate’s newest GPT-5 mannequin and increasing mannequin choices for paid subscribers.
In a put up on X, Sam Altman mentioned customers can now select between three response modes, “Auto”, “Quick”, and “Considering”, for GPT-5. Whereas most customers are anticipated to stay with the Auto mode, the extra settings goal to assist these searching for quicker replies or deeper, extra detailed reasoning, he mentioned.
For GPT-5 Considering, the weekly fee restrict has been set at 3,000 messages, after which customers may have entry to further capability by way of GPT-5 Considering mini.
The mannequin now helps a 196k-token context restrict, enabling it to course of for much longer conversations and paperwork. Altman famous that these limits might change over time primarily based on utilization patterns.
GPT-4o returns
He additionally confirmed that GPT-4o has returned to the mannequin picker for all paid customers.
“If we ever do deprecate it, we are going to give loads of discover,” Altman mentioned. Paid customers now even have a “Present further fashions” toggle in ChatGPT’s net settings, unlocking entry to choices like o3, 4.1, and GPT-5 Considering mini. The GPT-4.5 mannequin stays unique to Professional subscribers, with Altman citing its excessive GPU value.
Altman additionally revealed OpenAI is engaged on an replace to GPT-5’s character to make it “hotter” than its present tone however “not as annoying (to most customers) as GPT-4o.”
Final week, in its announcement, OpenAI mentioned GPT-5 was higher at coding and reasoning by way of advanced issues, and touted it as superior sufficient to show chatbot ChatGPT right into a PhD-level skilled.
GPT-5, unveiled throughout a livestreamed occasion on Thursday, is designed to be extra able to coding and artistic writing in addition to reasoning by way of advanced queries.
In a briefing with reporters this week, Sam Altman referred to as it a “main improve” from the corporate’s prior AI fashions. For “the primary time,” he mentioned, “it actually appears like speaking to an skilled in any subject.”
