OpenAI is working on improved memory, along with some minor enhancements. Recently, they shipped a build that includes new announcements related to memory improvements. The project, internally called “Moonshine,” will allow users to recall past conversations, expanding the scope of RAG for ChatGPT. In addition to memory, it will also reference previous chats more effectively. This feature will be available in Alpha mode, where users can choose to enable or disable it under the name “Improved Memory.”
Interestingly, this feature has been in development for some time. However, this week, Google released a similar recall feature for Gemini, which may have influenced OpenAI to accelerate their own timeline. Alongside memory improvements, OpenAI is introducing a new settings option allowing users to enable or disable follow-up suggestions.
There have also been traces of a new business plan, though details remain undisclosed. It is unclear how this plan will differ from existing offerings like Teams and Enterprise, but more information may emerge soon.
New ChatGPT web app updates
— Tibor Blaho (@btibor91) February 14, 2025
- mentions of a new ChatGPT Business plan
- new experiments for "word by word" and "block" response streaming animation
- updated announcement about Moonshine Alpha ("You're testing memory improvements: Let ChatGPT reference past chats, not just saved… pic.twitter.com/Pu3MLhdSxD
Over the past few weeks, there has been increased communication about upcoming features. One major development is the “Deep Search” feature, which is currently exclusive to Pro users. Soon, Plus users may gain access with up to 10 uses per month, while free-tier users might be allowed one or two searches per month.
Earlier this week, Sam Altman outlined OpenAI’s roadmap for model releases. He confirmed that GPT-4.5, previously known as “Orion” internally, will be launched soon. It is expected to succeed GPT-4o but will likely be OpenAI’s last non-reasoning model. Moving forward, GPT-5 will unify both general AI capabilities and reasoning functions.
OPENAI ROADMAP UPDATE FOR GPT-4.5 and GPT-5:
— Sam Altman (@sama) February 12, 2025
We want to do a better job of sharing our intended roadmap, and a much better job simplifying our product offerings.
We want AI to “just work” for you; we realize how complicated our model and product offerings have gotten.
We hate…
Notably, OpenAI will not be releasing the full version of o3 to the public. Instead, they are focusing on this unification, meaning that reasoning capabilities will be handled by a different model within GPT-5. As a result, the previously promised o3 full release has been postponed indefinitely, aligning with the broader GPT-5 launch timeline.
There has been speculation about release dates, with some predicting GPT-4.5 could arrive within weeks, while GPT-5 might take months. However, these estimates remain broad, and OpenAI has not provided concrete timelines.
Last year, OpenAI made several major announcements for ChatGPT in May, right before Google I/O. Given this pattern, it’s possible they will align their next major updates around the same timeframe.
