OpenAI releases GPT-5.5 Prompt, a brand new default mannequin for ChatGPT

On Monday, OpenAI launched a brand new basis mannequin known as GPT-5.5 Prompt, which can exchange GPT-5.3 Prompt because the default ChatGPT mannequin. The corporate mentioned the mannequin reduces hallucination in delicate areas equivalent to legislation, medication, and finance, whereas sustaining the low latency of its predecessor.

OpenAI launched the most recent GPT-5.5 mannequin final month with the corporate claiming enhancements in areas like coding and information work.

The brand new mannequin additionally achieved a rating of 81.2 within the AIME 2025 math take a look at, in comparison with 65.4 for the older mannequin. It additionally outperformed its predecessor on the MMMU-Professional multimodal reasoning benchmark, with a rating of 76.0 vs 69.2.

The discharge positioned a specific emphasis on context administration. GPT-5.5 Prompt can use its search software to refer again to previous conversations, recordsdata, and Gmail to provide you extra customized solutions. This function might be obtainable to Plus and Professional customers on the net, with plans to roll it out to cellular quickly. OpenAI mentioned that it plans to increase entry to this function to Free, Go Enterprise, and enterprise customers within the coming weeks.

With this replace, ChatGPT will salso how reminiscence sources throughout all fashions that will help you perceive the place it generated the solutions from. Customers can delete outdated sources or right them if the reply was mistaken. Crucially, the corporate mentioned that should you share a chat with somebody, they gained’t be capable of see the reminiscence sources.

For builders, the GPT-5.5 mannequin might be obtainable by means of API as ‘chat-latest,’ with 5.3 obtainable as an possibility for paid customers for under three months.

The corporate has confronted rebuttal from earlier mannequin withdrawal strikes. When OpenAI withdrew its GPT-4o mannequin, there was vital backlash from customers who related to the model’s “personality.” GPT-4o affirmed customers’ decisions often and that made them really feel a connection to that specific mannequin. Customers who signed petitions to cease OpenAI from retiring it described the mannequin as their “finest good friend” or “a mirror.” Regardless of the outcry, GPT-4o was deprecated in February 2026.